46 Commits

Author	SHA1	Message	Date
Michael Ramos	ed4f2e82fe	feat(review): Guided Reviews intro dialog + Guide button discovery hint (#1000) ... * feat(review): one-time Guided Reviews intro dialog + Guide button hint New GuideIntroDialog slots into the announcement chain (look-and-feel -> guide intro -> review setup, never stacking): a hero screenshot of a guided review with a brief plain-language description and a mock header strip pointing at the real Guide button. Shows once per user (cookie), and only when a Guide button actually exists (hasSearchableFiles) -- skipping the render does not consume the one-shot cookie. The hero is CDN-hosted at https://plannotator.ai/assets/guided-review.webp (committed to apps/marketing/public/assets so every marketing deploy re-ships it past the --delete sync; uploaded to S3 now so it is already live) instead of growing the single-file bundle; on load failure the image container hides entirely for a clean text-only dialog. Independently of the dialog, the header Guide button shimmers (TextShimmer) with a pulsing dot until the guide is first opened -- acked on ANY open path (button, shortcut, job auto-open), not just the button click. * fix(review): guide intro first in the dialog chain, full-width panel Chain is now guide intro -> look-and-feel -> review setup (was second); panel widened to max-w-5xl to match the look-and-feel dialog. Intro seen flag switched to the versioned-cookie pattern so future revisions can re-show it. * fix(review): latch guide-intro eligibility at first post-load render hasSearchableFiles changes on mid-session diff switches; an empty-to-non-empty switch would pop the intro over work in progress, or yank an open look-and-feel dialog (its gate reads guideIntroVisible). Eligibility is now decided once when loading clears -- dialogs only mount after that, so the latch is always set before they render. A skipped session keeps the cookie unconsumed. Also corrects AGENTS.md's chain order (guide intro first).	2026-07-05 09:49:00 -07:00
Michael Ramos	aa46dbb29a	feat: guide per-file summaries + GitHub Copilot CLI agent engine (#997) ... * feat(guide): per-file summary on guide diff refs Each diffs[] entry now carries a required (schema-enforced) 1-2 sentence summary of the semantic change in that file, written from the diff hunks alone. Sanitizer passes it through when it's a non-blank string and omits it otherwise -- a missing summary never drops the ref or fails the guide. Rendered as a muted inline-markdown line above each diff in the guide; marker-engine contract and demo data updated to match. * fix(guide): repair prompt must not invent missing summaries The guide schema now requires summary, so a schema-enforced repair of a payload that lacks them (marker-engine output) would force the model to fabricate captions with no diff in sight. Instruct it to fill missing required fields with an empty string instead; the sanitizer already renders nothing for blanks. * feat(agents): GitHub Copilot CLI as a marker engine for review + guide jobs Adds copilot as the fourth marker engine (no schema flag, so it uses the nonce-tagged marker-block contract like Cursor/OpenCode/Pi): - marker-review.ts: COPILOT_ENGINE — 'copilot --output-format json' JSONL stream (assistant.message carries the assembled text; deltas skipped), models discovered from 'copilot help config', live-log formatting for tool.execution_start/complete. Non-interactive posture: --no-ask-user auto-denies unallowed tools, --deny-tool=write, read-only-oriented shell allowlist (git/gh/glab/jj/wc), builtin MCPs and auto-update disabled. - Exported MarkerEngineId and replaced every 'cursor'|'opencode'|'pi' cast with it (Bun server, Pi server mirror, guide-review) so the next engine is a two-edit change. - agent-jobs (both runtimes): copilot in SERVER_BUILT_PROVIDERS; capability entry + model discovery come free from the MARKER_ENGINES loop. - UI: copilot review/guide engine with per-surface model settings (useAgentSettings), AgentsTab launch + config rows, GuideEmptyState launcher, job-detail labels, CopilotIcon (currentColor, official mark). - Tests: argv/read-only flags, help-config model parsing, stream reduction, full marker pipeline; profile-map expectations widened. Verified live: composed review prompt through the real copilot binary, marker block parsed, seeded bug found. * fix(agents): structurally deny high-consequence verbs for Copilot jobs git:*/gh:*/glab:* stay allowed for inspection ergonomics, but Copilot's deny-precedence rules now block the verbs a prompt-injected background job could abuse: git push/reset/clean/checkout/restore (local reviews run in the user's real working tree), and PR/MR/issue comment/create/merge/ close/edit/review on gh and glab. Probe-verified: git log runs, git push --dry-run is denied by the shell(git push) rule. Addresses the one confirmed finding from the PR #997 review round; the other two (untracked-file reads, git -C) were refuted by live probes.	2026-07-05 09:12:20 -07:00
Michael Ramos	d6b98f0c0d	feat(review): Guided Review + Pi agent-job provider (#993) ... * docs(adr): Guided Review ADR/spec/research + agent-provider fit studies ADR 006 (Guided Review as a first-class feature), four implementation spikes, synthesis, spec (iterated through preflight and review fixes), recap, and the Flue/Pi provider-fit research syntheses. * feat(review): Guided Review takeover + Pi agent-job provider Guided Review (ADR 006) — a Linear-Guides-style chaptered review: - guide agent-job provider (packages/server/guide/guide-review.ts): schema-constrained sections (title/overview/file refs) over the live patch, coverage-validated server-side (every changed file exactly once, fabricated paths dropped, fail-closed on empty output), importance-first ordering prompt with speed discipline (diff-first, no repo exploration) - runs on claude/codex natively and cursor/opencode/pi via the marker contract; guide-scoped low effort defaults for quicker generation - takeover UI (packages/review-editor/components/guide/): Guide header badge + Mod+Shift+G, full-width screen that CSS-hides (never unmounts) the file tree/dock, empty state with inline model pickers, skeleton loading, generating state with collapsed activity log, two-column section cards (sticky prose column with internally-scrolling file list, content-height diffs, review-independent collapse, peek semantics) - annotation parity: guide diffs mount the real DiffViewer with file-scoped handlers into the same CodeAnnotation state and feedback export; per-section reviewed state persisted via /api/guide/:jobId - reviewed/tour prompt speed sections; renderMarkdownProse shared between tour and guide (fenced-block support, muted tone variant) Pi agent-job provider: - third MarkerEngine (pi --mode json --no-session --no-approve) with live model-catalog discovery, thinking-level control (--thinking), fail-closed marker parsing (Pi exits 0 on in-run errors by design) - available for review + guide jobs from both launch surfaces Shared UX: - searchable, provider-grouped model pickers (auto over 12 options) - full Claude version catalog with latest-resolving aliases labeled - AgentControls primitives extracted from AgentsTab; cross-instance settings sync in useAgentSettings - Pi extension server hand-mirrors + vendor.sh entries throughout * fix(guide): PR-993 review fixes + failure recovery ladder Review fixes: - coerce marker-engine title/intent at ingest (prompt-enforced output could store non-strings and crash the takeover) - keep placed files out of unplacedFiles (exactly-once guarantee) - shared REVIEW_ENGINE_LABEL (fixes 'generated by pi' header; typed exhaustiveness prevents recurrence) - key ActiveGuide by jobId so per-guide focus state resets - formatModel handles marker-engine tour/guide jobs (empty model chip) - drop dead guideCodexFast; memoize estimateDiffHeight - store + display Pi thinking level on job cards Failure recovery ladder (auto -> one click -> manual -> regenerate): - mechanical JSON repair pass (fences/slice/trailing-commas/bracket balance) before any guide parse fail-closes - failed payloads captured per job (200KB cap, per-engine extraction) - 'Fix output' repair job: pure text transform on a schema-capable engine (claude/codex preferred), forced low effort, lands as a normal guide job; repairOf threaded through launch validation - editable output panel: GET /api/guide/:jobId/output prefills a textarea, POST /api/guide/:jobId/submit runs the same server-side validation and opens the fixed guide; inline errors for iteration - Pi extension mirrors + AGENTS.md endpoint/body docs * fix(guide): string-aware mechanical JSON repair (self-review) - terminate a dangling string literal before appending bracket closers (truncation mid-string is the most common shape; closers appended inside the string never parsed) - strip trailing commas outside string literals only (the regex could rewrite overview content, violating the repair contract) * fix(guide): launch-snapshot validation, codex strict schema, repair persistence PR-993 round-2 review fixes + live-test findings: - validate guide completion (and manual repair) against the LAUNCH-time changed-file set, snapshotted per job — switching diff/base/PR while a guide generates no longer destroys valid output (client already degrades stale refs per-file) - codex strict structured output: unplacedFiles now required in the guide schema (OpenAI 400s schemas with optional properties under additionalProperties:false — live-confirmed, then live-verified fixed: 5 sections / 83 files placed on a ChatGPT-account codex) - manual repair survives reload: successful /submit flips the job to done via completeJobExternally + status guard on the endpoint - codex model defaults -> gpt-5.5 (5.3-codex is deprecated; ChatGPT accounts reject it) with one-shot migration of saved picks; guide codex default reasoning stays low - pi extension guide branch snapshots launch state (TOCTOU hygiene) - paragraph collector no longer swallows an unspaced code fence - tour hook backports: out-of-order fetch guard; save outside updater - guide empty-state copy rewritten value-first * fix(guide): PR-993 round-3 — repair-diff fidelity, failure surfacing, pi tool hardening - repair jobs validate against the FAILED job's recorded file set (the payload under repair references that changeset; validating against the diff on screen at repair time re-introduced destroy-on-switch) - takeover follows the NEWEST running guide (progress + Cancel), and a failure newer than the shown guide surfaces as a dismissible strip (error + Fix output + Details) instead of being silently masked - pi jobs run with --exclude-tools edit,write (bash stays: the agent needs git inspection; full read-only would break generation) - tour/guide result ingestion fails closed on unexpected throw (done- looking jobs no longer 404 their result) - sidebar guide mode gated on file availability like the header badge - drop orphaned DEFAULT_GUIDE_CODEX_FAST (fast mode intentionally not offered for guide); retry owns its fetch cancellation; final trailing-comma pass after bracket-closing in the repair ladder; collapsed-row checkbox/expand are sibling buttons (a11y) * fix(guide): PR-993 round-4 — callout text, guide job detail, toggle races - single-line callouts (> [!IMPORTANT] message) keep their message — the form our own prompt solicits rendered an empty labeled box; type now derives from the [!TAG] capture, not a whole-line scan that mistyped '> [!NOTE] this is important' - job detail panel gets a guide status card (Open guide via a new ReviewStateContext.openGuide) instead of mislabeling 'Guide Generated' as a red Incorrect review - marker-engine guide jobs get readable live logs (engine fallback in the formatter lookup; provider stays 'guide') - repair prefers the failed job's own schema-capable engine (proven working on this machine) before falling back by binary presence - toggleReviewed/toggleChecked: pure functional updaters + persistence effect with seed-skip — race-proof AND StrictMode-safe (resolves the tradeoff previous rounds accepted) - dedups: formatDuration/ElapsedTime shared from AgentsTab; whichCmd exported in the pi extension; SEARCHABLE_THRESHOLD imported by InlinePicker - new renderMarkdownProse tests (single-line/multi-line callouts, paragraph passthrough) * fix(guide): PR-993 round-5 — keep engine-default model option, label Pi jobs - GuideEmptyState's marker catalogs mirror AgentsTab's per-engine semantics: opencode/pi prepend the engine-managed Default ('' value) to the discovered list (dropping it left saved-default users a blank pill with no way back); cursor replaces (its list includes 'auto') - job detail provider pill labels pi review jobs 'Pi' instead of the 'Shell' fallback * fix(guide): PR-993 round-6 — context scoping, stale models, repair engine, staging gate - guide takeover + auto-open scoped to the current review context: a guide launched against PR A no longer shows over PR B, and a guide finishing for an away context defers its auto-open until the reviewer returns to that context (unmarked in the dedupe set on purpose) - guide launcher reconciles saved cursor/opencode/pi model ids against the live catalog at read time (picker + launch share one effective value) — no more posting dead ids after an account switch - repair prefers the failed job's OWN engine whenever its binary is present (provably runnable; marker binaries resolved via MARKER_ENGINES — cursor's CLI is 'agent'); claude/codex are fallback only, killing the broken-claude repair doom loop - per-file staging gate (canStagePath) in guide diffs AND the pre-existing same gap in ReviewDiffPanel — committed-only files in since-base reviews no longer offer a no-op Git Add * fix(review): scope tour auto-open to current context (self-review) Same cross-context gap just fixed for guides: a tour finishing for PR A popped its dialog over PR B. Both auto-open effects now share one jobMatchesCurrentContext helper with the same deferred-open semantics. * fix(guide): PR-993 round-7 — complete the context-scoping story Round 6 scoped guides to their review context; this closes the two surfaces that scoping left dangling, plus a focus gap: - switching to a context that already HAS a completed guide now shows that guide: the takeover falls back to the context's newest done guide job when activeGuideJobId belongs elsewhere (previously landed on the empty state with the guide sitting unreachable) - 'Open guide' affordances are context-gated everywhere they exist: job cards hide it for cross-context guides (opening can't switch PRs), and the job detail panel explains where the guide belongs instead of offering a dead button; one shared jobMatchesReviewContext predicate now backs App, GuideScreen, AgentsTab, and the detail panel - file-chip navigation retargets the guide's focus arbiter (scrolling under a stationary pointer fires no pointerenter, so the annotation toolbar stayed bound to the previous diff) * fix(tour): retry owns its fetch cancellation; document prUrl-only scoping - useTourData retry converts to the nonce-driven effect re-run pattern (the round-4 backport carried the fetch guard but not the retry fix useGuideData got in the same batch — half a backport) - jobMatchesReviewContext now documents WHY it matches by prUrl only and not diffScope/diffContext: guides/tours reference files and degrade per-file when the diff shifts; scope-strict matching would hide useful artifacts on layer/base/mode switches. Deliberate, do-not-tighten-without-UX-decision * fix(guide): PR-993 round-9 — worktree-aware context, guide-scoped marker settings - jobMatchesReviewContext now compares diffContext.worktreePath for local jobs: prUrl + worktree define WHERE the review is (matched strictly); mode/base/scope remain WHAT VIEW (deliberately loose) — a guide launched against worktree A no longer opens over worktree B. Jobs predating the snapshot coalesce to the main tree. - Cursor/OpenCode/Pi get guide-scoped model/thinking settings (guideCursor/guideOpencode/guidePi), mirroring the existing guideClaude/guideCodex isolation: tuning a guide's marker model no longer silently changes the next code review with that engine. Defaults are each engine's natural default, not seeded from review settings (same precedent as guideClaudeEffort='low'). AgentsTab's reconcile effects snap both surfaces against the shared catalogs. * fix(review): keep worktree parsing browser-safe (round-9 build fix) parseWorktreeDiffType lives in shared/review-core, which imports node:path at module top — pulling it into App.tsx broke the vite browser build (typecheck and tests don't catch it; only the bundle does). Context matching now reuses App's existing hand-parsed activeWorktreePath memo — one parse, and it's the same one that drives the sections/tree UI, so matching aligns with what's on screen. Also fixes a deps-array reference to the removed memo that bundled fine (treated as a global) but would have thrown a ReferenceError at runtime. * docs(review): mark worktree-subtype lockstep between server parser and App copy * fix(review): PR-993 round-10 — guide shortcut gate, settings sync race, heading markdown - bare a/v staging/viewed shortcuts suspend while the guide takeover is open (the dock is only CSS-hidden, so a diff panel stayed 'active' underneath and bare keys acted on an invisible file) - cross-instance settings sync replaces the was-last-update-remote boolean with value comparison (lastSyncedJsonRef): the flag conflated 'a commit happened' with 'the last change was remote', so a local edit batched with an incoming broadcast silently skipped its own cookie write and rebroadcast - ### headings in guide/tour prose run through renderInlineMarkdown like every other block (backticked symbols rendered as raw tokens); regression test added * fix(guide): PR-993 round-11 — reveal channel for sidebar jumps, repair exit code - new guideRevealFile channel closes the documented round-7 gap and its AI sibling: sidebar jumps (annotation clicks, AI line citations) made while the guide takeover is open no longer mutate the hidden dock's active file; instead the GuideSectionCard containing the target file expands its collapsed (reviewed) section, focuses the diff, and scrolls to it — so jumps into collapsed sections stop silently no-opping. Cleared on guide close/switch so keyed remounts don't replay the last reveal. - completeJobExternally resets exitCode to 0 (both servers): a successfully repaired guide kept showing the failed run's Exit 1 chip in the job detail panel * fix(guide): clear reveal channel in-batch at guide-switch sites (self-review) The clear-on-change effect fires after a switched guide's keyed cards have already mounted (child effects run before parent effects), leaving one commit where a stale reveal from guide A could expand+scroll a same-named file's section in guide B. All three setActiveGuideJobId sites now clear synchronously in the same batch; the effect stays as a backstop for close and future set sites. * fix(guide): PR-993 round-12 — titled sections, honest large-PR validation - sanitizeGuideSection gives every surviving section a non-empty title ('Untitled section' fallback): a diffs-only section rendered as a blank chapter inside a 'Guide Generated' job — no parse failure, so no recovery flow. Keeping (titled) beats dropping: placed files are not in unplacedFiles, so dropping would orphan them silently. - guide launches on large PRs (layerPatchIncomplete) recompute changedFiles from the local checkout (git diff --numstat origin/<base>...HEAD, rename/binary handling) when available — the PR-mode prompt tells the agent to read that full local diff, but the changed-files block and validation snapshot came from the truncated platform patch, under-listing files to the model and then dropping its valid refs (or failing the guide) at validation. Falls back to the partial list on any failure. Mirrored in the Pi extension server. Round-12's P1 (Pi marker extraction reads message_end, 'should' read text_delta) was refuted by live probe: Pi's assistant message_end carries the fully-assembled content blocks, which is exactly what piExtractText reads; the delta-reading code the finding cross- referenced is the Ask-AI STREAMING provider, a different job. * test(guide): first direct coverage for guide-review's pure logic (self-review) The module's repair ladder, validation, and stream parsing had zero unit tests — exercised only end-to-end through live agent runs. Pins the behaviors the review rounds fixed: blank-title fallback (round 12), first-placement-wins dedup, changedFiles filtering + fail-closed empty guides, prose-only vs lost-diffs section handling, unplacedFiles merge/dedup/fabrication-filtering, title/intent coercion, trailing- comma and unbalanced-bracket repair, truncated stream-line recovery. 13 tests; server suite 374 → 387. * fix(guide): PR-993 round-13 — full-stack recompute regression, flag snapshot, unplaced reveal - the round-12 large-PR recompute now runs ONLY in layer scope: in full-stack scope launchPatch is already a local full recompute, and the layer diff (origin/<base>...HEAD) was the WRONG file set — it omits earlier stack layers' files, dropping their refs at validation or failing the guide. Introduced by round 12; caught before any release. Both servers. - layerPatchIncomplete is snapshotted with the other launch locals (launchLayerPatchIncomplete): coherence with launchPatch was positional (same sync segment) — now structural, so a future await inserted upstream can't silently desync them. Both servers. - the 'Everything else' bucket handles guideRevealFile: sidebar jumps to unplaced files now focus + scroll their diff (the section-card effect only covers placed files) — closes the residual gap noted in the round-11 self-review	2026-07-04 10:23:59 -07:00
Michael Ramos	e8df06db7c	feat(review): Commits panel — linear history rail with per-commit diffs (#994) ... * docs(adr): spec for the commit-list review view (linear history rail) * feat(review): commit:<sha> diff mode along the shared diff-type seams A new git-only diff family for reviewing one historical commit against its first parent (git-show style), threaded through every seam a diff type crosses: - runGitDiff: git diff <sha>^ <sha>; a root commit diffs against the empty tree. Label is 'Commit <shortsha> — <subject>'. The sha is validated as bare hex (parseCommitDiffType) before it reaches any argv position, on top of --end-of-options. - getGitDiffFingerprint: anchored to the sha alone (present/gone), NOT headSha — new commits landing mid-review don't change this diff and must not raise the staleness banner. A vanished commit (rebase + gc) flips present→gone and fires it honestly. - getFileContentsForDiff: old side <sha>^ (null on root), new side <sha>. - parseWorktreeDiffType learns worktree:<path>:commit:<sha> (the sub-type contains a colon, so it needs its own split) — commit review composes with worktree sessions like every other mode. - vcs-core: the git provider owns commit:*; staging stays gated off (canStageFiles allowlist unchanged — nothing in a historical commit is stageable). - Ask AI context: a commit-mode inspect string ('git show <sha>', with an explicit 'historical commit, not the working tree' note). Also adds listCommitHistory for the upcoming /api/commits endpoint: one --first-parent page from HEAD (before=<sha> continues at its first parent, +1 fetch for an honest hasMore), with isHead / isRepoUser / isPastBase flags. isPastBase is reachability from the base — computed as the complement of 'rev-list --first-parent HEAD ^base', which is a prefix of the walk, so the client's single divider is exhaustive. An unresolvable base degrades to no divider (matches since-base's posture on such repos). * feat(review): GET /api/commits — linear history endpoint (Bun + Pi) One page of the branch's --first-parent history (?limit=&before=), served by listCommitHistory against the active diff's cwd (worktree- aware via resolveVcsCwd) and the active base (so the divider tracks the same baseline the review compares against, including the startup origin/<default> upgrade). Gated to plain local git sessions — PR / workspace / jj / p4 return 400, mirroring the client's commitsCapable gate. Both runtimes; the route-parity test covers the pair. commit:<sha> switching needs no endpoint changes: /api/diff/switch already dispatches through runVcsDiff (the git provider owns the new family), the epoch guard applies as-is, the sections sidecar correctly stays since-base-only, staging 400s via the canStageFiles allowlist, and baseRelevantDiffType keeps the behind-GitHub banner suppressed for commit diffs. * feat(ui): Commits panel view — the linear history rail Third left-panel view: 'Git status | Commits | Tree'. The Commits view is a pure commit list (short sha, subject, relative age compacted to '2h', author only when it isn't the repo user, HEAD dot, a divider where the branch meets the resolved base, 'Show more' paging). It never becomes a file list: clicking a commit switches the diff to commit:<sha> and the existing needsInitialDiffPanel flow lands the center dock on the all-files surface; re-clicking the active commit just re-focuses that panel. Persistence: reviewPanelView gains 'commits' with NO diff coupling — the view persists, the diff opens on the user's normal default until a commit is clicked, and a sha is never persisted (it may not survive a rebase). The sections⟺since-base coupling is untouched, but the classic-diff→Tree snap in Settings/ReviewSetupDialog now only fires when the current view is 'sections', so picking a default diff no longer stomps a Commits preference. The load-time self-heal ignores 'commits' by construction (it keys on panelView === 'sections'). Gates: commitsCapable = plain local git session (no PR / workspace / jj / p4, matching the server's /api/commits gate); the segment is hidden elsewhere. Staging is inert in commit diffs on both ends (STAGEABLE_DIFF_TYPES and the server allowlist never match commit:*). Worktree sessions compose — the client parses worktree:<path>:commit:<sha> and handleDiffSwitch prefixes as usual; the rail refetches on worktree/base changes but deliberately not on commit clicks, so paging state survives selection. Accepted edges (v1): no j/k selection in the rail (each move would run a full diff switch); toggling Commits→Tree keeps a commit diff active, where the tree's diff-type dropdown marks nothing — the gitRef header still names the commit. * fix(review): self-review cleanups on the commit-list feature - listCommitHistory: a repo with no commits yet (unborn HEAD) returns an empty page instead of null — the panel showed a 500-backed 'Retry' error where 'No commits' is the truth. Every other review surface already degrades gracefully on a commit-less repo; now this one does. Only a genuinely unanswerable repo (not a repo at all) stays null. - Hoist the bare-hex sha rule into one BARE_HEX_SHA_RE shared by parseCommitDiffType and the before-cursor validation (was written twice). - The 'all' diff case now uses the getEmptyTreeSha helper the commit mode extracted, instead of keeping its own inline copy. - Drop CommitsPanel's isLoadingDiff prop — declared and passed but never used. * feat(ui): commit cards, labeled groups, avatars — review round 1 Four items from the live review round: - Toggle: Commits moves to the far right (Git status | Tree | Commits) and the segment text goes 10px → 12px (padding trimmed to keep three segments inside the 256px header). - The bare '── origin/main ──' divider confused more than it explained. Replaced with two labeled groups: 'On this branch' (commits not yet reachable from the base) and 'In <base>' (shared history), each with a tooltip spelling out what the boundary means. - Rows become overview cards: subject (2-line clamp) on top; a meta row with author avatar, author name (only when it isn't the repo user), a HEAD badge, short sha, and compact age. Active card gets the primary border/tint. - Author avatars, reusing the PR machinery rather than reinventing it: parseRemoteUrl/parseRemoteHost classify the origin forge; the gh/glab invocation shape (incl. the --hostname self-hosted convention) and GitLab's relative-avatar absolutization rule are mirrored from pr-github/pr-gitlab. What could NOT be reused directly: PR avatars are keyed by platform login, which local commits don't have — so GitHub resolves via one repos/{o}/{r}/commits?per_page=100 call per session building an author-EMAIL → avatar map (unpushed commits by the same author resolve through it; verified live), and GitLab uses its per-email /avatar endpoint (capped, deduped). Everything is memoized per session and fails closed to the initials fallback — a repo with no remote, an opaque self-hosted host, or an unauthenticated CLI never delays or errors the commits endpoint. The Avatar component moves out of PRCommentsTab into its own module, now shared by the PR timeline and the commit cards. New shared module commit-avatars.ts is vendored to Pi (vendor.sh) and the enrichment runs in both runtimes. CommitListEntry gains authorEmail (resolver key) and avatarUrl (server-enriched). * ui(review): flush commit rows, prominent base boundary — review round 2 - Cards are gone: rows stack flush (no borders, no margins), hover tint only. Subject is strictly one line, ellipsized (no clamp/wrap). - Author name always shown (was: only when it differed from the repo user), with the avatar, on the compact meta line under the subject. - The base boundary is a prominent labeled rule ('— In origin/main —', foreground-weight lines) instead of the faint border under a group header; the 'On this branch' header above stays quiet. * docs(adr): evaluate local stack-parent detection — scenario matrix, no build decision Empirical stress-test of the merge-base-reachability algorithm across 9 scenarios plus this repo itself. Findings: the stacked-ness test is graph truth (never lied); parent attribution has one content-harmless failure (same-merge-base sibling labels) and one dangerous, provably graph- undetectable one (branches pointing inside B's own history win and silently hide the user's commits). Determination: only viable as a suggest-and-confirm affordance with per-branch persisted choice — never silent auto-defaulting. Decision to build deliberately left open. * fix(ui): Commits view is session-only — never the opening view Live-review catch: the panel toggle inherited the sections/tree 'last choice becomes the default' cookie write, so clicking Commits in one session made the NEXT review open on the history rail instead of the diff. Ruling: a review always opens on the persisted sections/tree default; Commits is entered via the toggle each session. - reviewPanelView narrows back to 'sections' | 'tree'; a stale 'commits' cookie is treated as unset (self-heals to the default). - The live view is now a session overlay in App state: selecting Commits flips it without touching config; selecting Git status/Tree clears it and persists as before. Single choke point in selectPanelView, so every toggle site inherits the rule. - Settings loses the Commits segment (it configures the OPENING view, which Commits can no longer be). Spec updated with the revision. * ui(review): the panel toggle is session-only — it never writes the default Second live-review ruling on persistence: the header toggle shouldn't persist ANY choice, not just Commits. Looking at another view mid-review must not silently change what the next review opens on. - selectPanelView becomes a pure session override (local state layered over the persisted value); no configStore writes from the toggle path. - handleSwitchToSections drops its defaultDiffType write — that write only existed to keep the persisted view/diff pair consistent, and the toggle no longer persists a view. Settings and the setup dialog remain the only reviewPanelView/defaultDiffType writers and keep enforcing the sections ⟺ since-base coupling; the load-time self-heal still repairs conflicted pairs those writers may leave. - Spec + settings-registry comments updated to name the new rule. * feat(ui): commit description card + collapsed all-files for commit diffs Clicking a commit now answers 'what and why' before 'which lines': - New commitInfo sidecar (Bun + Pi, same mode-conditional shape as sections): when a commit:<sha> diff is active, /api/diff and /api/diff/switch carry the commit's full metadata — subject, multiline body, author (+ avatar via the session resolver), sha, age — from a single 'git show -s' (getCommitDiffInfo in review-core, body as the trailing format field so newlines survive). - CommitDescriptionHeader heads the all-files surface: subject, author row, and the full body rendered as markdown via the same MarkdownBody the PR viewer uses. Long bodies scroll inside the card (max-h) so the diff keeps the viewport. - Commit diffs open FOLDED: AllFilesCodeView gains defaultCollapsed — items seed collapsed at build time, the seed is part of fileSetKey (CodeView seeds once per instance, so a seed change must remount), and the collapse-all toggle state initializes to match. Every other diff mode keeps opening expanded; leaving the commit diff clears the card and the folded default together (the sidecar is absent). Verified live through the compiled binary: switching to a real commit returns subject/body/avatar, switching back to since-base clears it. * ui(review): commit description scrolls with the diff — no pinned card Round feedback: the fixed header with its own inner scrollbar felt embedded; the description should read as the top of the document and scroll away with the files. - AllFilesCodeView gains leadingContent: rendered via a portal INTO CodeView's scroll container (absolutely positioned at content top, so it participates in the scrollable overflow), with its measured height fed into layout.paddingTop so the virtualized items start below it. CodeView stays the single scroll authority — no nested scrollers, no wrapper flex row. - CommitDescriptionHeader shows the full body inline (no inner OverlayScrollArea). Only genuinely huge bodies (>24 lines / >1800 chars) get a CSS max-height clamp with a fade mask and a Show more/Show less toggle — CSS clamping keeps the markdown intact, and the ResizeObserver feeds the expanded height back into paddingTop. Keyed by sha so the toggle resets per commit. * ui(review): entering the Commits view auto-opens the HEAD commit Toggling to Commits used to leave the previous mode's all-files diff on screen — a rail full of commits next to content that belonged to none of them. Now entering the view selects the HEAD commit (top of the rail) so the center immediately shows its diff + description card. Once per entry, ref-guarded: an already-active commit diff survives a toggle round-trip, a user click supersedes it, and a failed switch doesn't retry-loop. When entry races the first log fetch, the effect fires as soon as the log lands. * perf/ui(review): fast commit navigation — instant veil + skip context recompute Rail clicks felt laggy and jumpy: the old diff sat on screen for the whole switch round-trip and then snapped, and the switch itself was doing a full getVcsContext recompute (branch + worktree + recent-commit enumeration — three git walks) that a commit click can never invalidate. - Same-cwd commit:<sha> switches skip the context recompute in both runtimes (the client keeps its existing context when the field is absent — long-standing contract). Measured through the compiled binary: commit switches now ~40-60ms server-side. - The center dock shows an immediate 'Loading commit…' veil whenever a commit switch is in flight (or the Commits view was just entered and HEAD auto-select hasn't landed) — the stale previous diff never shows and the click reads as instant. Errors surface through the normal empty-state, never trapped under the veil. * fix(ui): PR-994 review round — toggle/settings separation + commit-nav fixes (App) The headline fix: the load-time settings repair guarded on the live panelView, which now carries the session toggle override — so clicking 'Git status' once mid-review (saved prefs tree + a classic diff) hit the healer and silently persisted defaultDiffType='since-base'. The toggle must never be a settings writer, directly or through a repair path. Keyed to persistedPanelView now; audited the remaining configStore.set call sites — no toggle path can reach a write. Also in this file, same review round: - Commit-navigation veil gains a real predicate: it drops on a commit-log fetch error (rail shows Retry; center was stuck under the spinner forever) AND on a genuinely empty history (zero commits — the second stuck-veil trigger the automated review missed), while still covering the log-loading and pre-auto-select frames. - handleSelectCommit composes the worktree prefix once and calls fetchDiffSwitch directly (the equality check and the switch previously used two different composition paths). - commitsCapable/showCommitsPanel hoisted above the global keyboard handler; Cmd+F is no longer intercepted in the Commits view (no search input exists there — the browser find takes over instead of a silent no-op). * fix(review): PR-994 review round — commit-log staleness, loading flags, avatar cap - useCommitLog clears its cached list when the history context changes (worktree/base switched while the view was away): the HEAD auto-select acted on the stale rows and opened the previous context's commit. Same contextKey still keeps the cache (no empty flash on re-entry). - The hook's cleanup now resets both loading flags: a generation-skipped finally never cleared them, leaving 'Show more' stuck disabled as 'Loading…' after leaving mid-page. - commit-avatars memoizes misses only for emails an attempt actually covered — the GitLab per-call cap (10) was recording every email past it as permanently unresolvable for the session. Regression test added. - GitLab host detection tightened from a bare substring to a structured match (gitlab.com / gitlab.* / *.gitlab.*) — 'mygitlabproxy.example.com' no longer classifies. - Collapse-all mirror re-derives from live item state after per-file toggles: with commit diffs seeding all-collapsed, expanding one file left the dock button on 'Expand all', and clicking it re-collapsed the file the user just opened. - The commit description card element is memoized on commitInfo identity so the measuring ResizeObserver stops churning on every context render. * fix(ui): entering the Commits view ends any open search session (self-review) The Cmd+F guard from the review round only blocked NEW searches in the Commits view; a search opened in Git status/Tree survived the toggle as hidden-but-live state — the query kept matching against the commit diff, marks kept rendering in the all-files body, Enter/F3 kept stepping matches, and the input to see or edit the query didn't exist anywhere on screen. handlePanelViewSelect (the single choke point every toggle site routes through) now clears and closes search on entry to Commits. Also re-traced the rest of the round under self-review — veil terminal states (log error/empty/loaded, switch failure, background refresh with a commit active), the hook's cleanup-before-rerun flag ordering, the avatar attempted-set on broken platforms, and both collapse-mirror paths — no further findings. * fix(ui): PR-994 round 2 — live rail freshness, non-destructive errors - The rail now keeps itself fresh: while the Commits view is visible, a quiet 10s poll head-compares page 1 and adopts it only when history actually moved. The commit DIFF's sha-anchored fingerprint stays as designed (a historical commit never goes stale, so the banner correctly stays quiet) — but the rail no longer freezes while an agent commits in the background. Adoption bumps the fetch generation so an in-flight 'Show more' from the old history can't append stale rows, and transient poll failures disturb nothing. - A page-1 fetch resets isLoadingMore: a refresh superseding an in-flight paging request skipped that request's generation-guarded finally, leaving 'Show more' stuck on 'Loading…' — rare before, a live race once the poll exists. - Errors are non-destructive when a list is on screen: a failed page or background refresh renders as an inline row with Retry under the list instead of replacing the whole rail; the full-panel error state is reserved for an empty rail. * perf/refactor(review): parallel GitLab avatar lookups; drop dead isRepoUser - The ≤10 per-call GitLab avatar lookups run in parallel instead of sequentially — they sit on /api/commits' critical path, and serial subprocess spawns added seconds to the rail's first paint on multi-author GitLab histories (PR-994 round 2 nit). - CommitListEntry.isRepoUser removed: v1 showed the author only when it wasn't the repo user; the live-review pivot to always-shown names left the field computed (one git config subprocess per page), serialized, and tested but never read, with a doc comment describing behavior the panel no longer has. Spec updated to match the revised row design. * fix(ui): commit-poll adoption clears all superseded fetch state (self-review) Adopting a new history from the background poll bumps the generation, which strands any in-flight fetch's generation-guarded finally — the same stale-flag class just fixed for isLoadingMore, but for isLoading (slow first load overtaken by the poll) — and a lingering inline error from the replaced history would otherwise sit under the fresh list. Adoption now clears isLoading, isLoadingMore, and error together. * fix(ui)/docs: PR-994 round 3 — worktree switches drop commit diffs; API docs - handleWorktreeSwitch treats commit:<sha> as non-portable: every other diff mode recomputes meaningfully against the target worktree, but a commit diff is context-bound content (worktrees share one object database — the reviewer's claimed git error doesn't exist, verified empirically), so 'preserving' it just re-rendered the OLD context's commit byte-for-byte. It now falls back to the session default (same option-availability rule resolveInitialDiffType applies). Swept every other activeDiffBase/diffType carrier: the whitespace toggle, staleness refresh, and fetch-base deliberately recompute the SAME diff (valid for commits); handleBaseSelect and the base-picker row already exclude commit mode; job diff-context labeling and the guarded diff-type dropdown are display-only. This was the single leak. - AGENTS.md (CLAUDE.md symlink): /api/commits row added to the Review Server table; /api/diff and /api/diff/switch now document the commitInfo sidecar and the commit:<sha> diffType family; the since-main section describes the three-segment session-only toggle and the never-persisted Commits view. * docs: normalize arrow glyphs in the /api/commits table row (self-review) * fix/refactor(review): PR-994 round 4 — rebase-safe paging, locale-proof ages, shared parsing - Pagination is rebase-safe: a 'Show more' cursor from a history that was rewritten mid-session (rebase/force-push) still resolves in the object store but is no longer on the branch — paging from it walked the orphaned pre-rewrite chain for the ≤10s window before the freshness poll adopts the new history. listCommitHistory now ancestor-checks the cursor (merge-base --is-ancestor) and returns an empty terminal page for orphaned AND vanished cursors alike; the poll replaces the list moments later. Regression test rewrites history between pages. - Ages are locale-proof: git localizes %cr via gettext ('vor 2 Stunden'), which the English-only compactAge regex silently couldn't shorten. CommitListEntry/CommitDiffInfo now carry committedAt (epoch ms, %ct); the rail and description card format it with the existing formatRelativeTime — compactAge and the third formatter variant are gone. - The %x1f over-split repair (fixed head/tail fields, rejoined free-text middle) was triplicated across listRecentCommits, listCommitHistory, and getCommitDiffInfo; one splitCommitFormatFields helper (head/tail counts, tail=0 covers the multiline-body shape) owns the edge case. - Avatar resets its broken-image state when src changes — latent today (every caller keys by identity), but it's a shared component and must be safe for callers that update src in place. Left alone from this round: hashString/hashFingerprintPart duplication is pre-existing, and the suggested cross-package import would drag node:path into the browser bundle — the documented reason client mirrors exist. * fix(review): PR-994 round 5 — merge-commit prompts, boundary-aware poll, auto-select guard - Agent prompts for commit:<sha> diffs instruct the exact on-screen command: git diff <sha>^ <sha> (first parent), explicitly steering away from git show — whose combined-diff presentation for a MERGE commit renders a different (often empty) changeset than the one under review. Root-commit fallback noted. One site covers Ask AI and every launched reviewer (and Pi via vendoring); prompt test pins the command. - The rail's freshness poll now adopts on boundary/base movement, not just a new head: an agent running git fetch advances origin/<base> while HEAD stays put, which re-partitions isPastBase — previously the head-only compare skipped adoption and the 'In origin/main' divider stayed stale until the view was reopened. Boundary is compared over the page-1 overlap window (a divider paged deeper than the probe can see re-syncs on the next full reload — accepted micro-edge). The reviewer's proposed trigger (the Fetch banner) is unreachable in commit mode; the external-fetch trigger is the real one. - The HEAD auto-select never fires while any diff switch is in flight — hardening only: the claimed first-click overwrite isn't reachable (isLoadingDiff was never a dependency of that effect, and auto-select settles before a human can click), but the guard makes the invariant explicit. * docs(adr): spec the pre-merge commits-view structure refactor Three behavior-preserving moves scoped for PR #994 before merge, motivated by the audit finding that every recent review-round bug lived at the App.tsx <-> useCommitLog seam: (R1) unify the commits-session state machine (poll, auto-select, veil) in one hook; (R2) lift the commit-rail block out of the 1.6k-line review-core into commit-history.ts; (R3) share the isSameCwdCommitSwitch predicate both runtimes inlined. Ownership tables name what deliberately stays put; execution runs smallest-first with full gates per commit. * refactor(review): share isSameCwdCommitSwitch across runtimes (spec R3) The ~10-line parse-and-compare predicate both /api/diff/switch handlers inlined moves to review-core beside its inputs (parseWorktreeDiffType / parseCommitDiffType — the canonical home for diff-type string logic); both call sites collapse to one line. Unit tests cover plain, worktree, cross-worktree, and non-commit-target cases. Behavior identical. * refactor(shared): lift the commit-rail block into commit-history.ts (spec R2) review-core.ts (1.6k lines) was absorbing the ~230-line Commits-panel data layer it doesn't need to own: CommitListEntry/CommitHistoryPage/ listCommitHistory and CommitDiffInfo/getCommitDiffInfo move verbatim to a new commit-history.ts, matching the package's per-concept split (jj-core, pr-stack, commit-avatars). The commit:<sha> DIFF plumbing (parseCommitDiffType, the runGitDiff/fingerprint/file-contents cases) stays in review-core with the other diff types — it participates in the dispatch; the rail does not. Wiring: review-core exports its three parsing primitives (COMMIT_FIELD_ SEP, splitCommitFormatFields, BARE_HEX_SHA_RE — listRecentCommits still uses them, so they can't move); package exports + vendor.sh gain the module (vendored siblings import relatively, same as pr-provider → pr-github); types.ts re-exports split by source; both runtimes import from the new module. History/metadata test suites move to commit-history.test.ts with the package-style per-file git harness; commit-DIFF tests stay in review-core.test.ts. Behavior identical — verified live through the compiled binary (/api/commits + commit switch + commitInfo sidecar). * refactor(ui): useCommitsView owns the whole commits-session machine (spec R1) The list cache, freshness poll, HEAD auto-select, and center-dock veil were split between useCommitLog and App.tsx — and every sync bug found across three review rounds (stuck flags, stale auto-select, adoption races) lived at exactly that seam. The auto-select effect and the veil derivation move into the hook verbatim (renamed useCommitsView, git mv), so the machine's invariants are locally checkable in one file. App now supplies only what it owns — visibility, the active commit, switch state, and onOpenCommit (its handleSelectCommit: the SAME path user clicks take, so auto- and user-selection cannot diverge by construction) — and consumes panel props plus veilActive. Net ~45 lines of App.tsx's most delicate logic deleted. One equivalence note: the hook's single "enabled" (showCommitsPanel && origin) now gates auto-select/veil where App used bare showCommitsPanel; the two only differ when origin is unset, which implies no gitContext (demo mode) and therefore showCommitsPanel === false — identical in all reachable states. Deliberately still in App (each guards a flow App owns): capability gating, handleSelectCommit, the Cmd+F / worktree-fallback / staleness-refresh / search-clear guards, and render wiring. Behavior identical; spec: adr/specs/refactor-commits-view-structure-20260703.md.	2026-07-04 10:13:11 -07:00
Michael Ramos	795f381ebe	feat(review): "All changes" git-status review view (#990) ... * feat(review): "Since main" git-status review view + correctness fixes Adds a composite `since-base` diff — merge-base(origin/main, HEAD) vs the working tree, plus untracked — as the default code-review view, rendered as a three-section "git status" panel (Committed / Changes / Untracked) with a Sections|Tree toggle, a first-run setup chooser, and a "baseline behind GitHub" fetch banner. Everything normalizes to one git patch, so the diff viewer, annotations, and feedback path are unchanged. New diff type wired through runGitDiff / fingerprint / file-content / context / staging / agent-context in packages/shared + both server runtimes (Bun + Pi mirror). Default flipped from `unstaged` to `since-base`; the old DiffTypeSetupDialog is replaced by ReviewSetupDialog (view + default-diff chooser, screenshots, Settings access). Includes a reviewed batch of correctness fixes: - P0: baseBehindRemote was permanently true (rev-parse missing --verify) - fingerprint blind to quoted/unicode untracked paths (unquote) - sections parser: record both sides of a rename; staged-delete wins over untracked (rm --cached collision) - graceful degrade when merge-base can't resolve (trunk/no-remote repos) - /api/fetch-base re-queries remote (narrow-refspec honesty) - /api/diff/switch concurrency guard (diffSwitchEpoch) + draft rekey - decouple remote-staleness probe from initialBase (Pi parity) - keyboard file nav in the sections view; banner gated to base-relative modes See adr/decisions/005-since-base-github-view-default-20260701-223706.md * fix(review): since-base review-round hardening (11 fixes, both runtimes) Addresses the multi-agent + PR-990 review findings: - Sections rename parser split ` -> ` on the raw quoted porcelain token, so a filename containing ` -> ` tore into garbage and a dirty file could render as "Committed". Now quote-aware (splitPorcelainRename + unit tests). - /api/diff/switch captured the epoch AFTER await req.json(), letting a slow-body older switch overwrite a newer confirmed one. Epoch now captured before any await; hideWhitespace committed only on win. Both runtimes. - Unresolvable base (trunk / no origin/HEAD) no longer auto-defaults to a degraded since-base that hides committed work — getGitContext only offers since-base when the base ref resolves, so the default falls through to uncommitted. Fingerprint + file-content degrade to HEAD to match the diff. - git rm --cached no longer yields two diff entries for one path (untracked files already in the tracked patch are dropped) — fixes wrong-file-open and j/k nav looping in the Sections panel; also fixes latent uncommitted/unstaged. - Settings' "Default Diff View" list now preserves the sections<->since-base coupling (can't leave an invalid sections+classic pair). - "Behind GitHub" banner only treats origin/* as fetchable; a bare local base ("main") is upgraded to its tracking ref at startup so Fetch can clear it. - hashUntracked resolves untracked paths against the repo toplevel, not cwd, so a review launched from a subdirectory isn't blind to untracked edits. - Agent review instruction now tells agents to enumerate/inspect untracked files. - Terminology: one "Committed changes" label; the live "Since <base>" label is dynamic (matches the header); "Since main" kept only as product copy. - Docs: CLAUDE.md endpoints/fields + since-base view section; ADR recap. Verified: typecheck (all projects), bun test 1793 pass, and empirical repro of the rm --cached dedup, trunk->uncommitted fallback, and rename parser. * fix(review): restore "(PR view)" on the Committed-changes label (unify toward it, not away) * fix(review): close since-base coverage gaps from PR-990 review round 2 - First-run setup no longer forces since-base on repos where it isn't available (base ref can't resolve). getGitContext omits since-base there; the first-run block now checks the same availability before resetting the default or showing the chooser, so committed work isn't silently hidden on trunk/no-origin repos. - reviewBase now includes 'since-base', so changing the base while in the git-status view passes the selected base to /api/file-content — expandable context is fetched from the right merge-base (was falling back to default). - Settings now mirror ReviewSetupDialog's coupling exactly: Sections ⇒ force since-base; Tree ⇒ leave the diff (Tree + since-base is valid and now saveable); classic diff ⇒ Tree; since-base diff ⇒ leave the view. Removes both coercions that previously discarded a supported preference. * fix(review): PR-990 review round 3 — fingerprint/dedup/base-canonicalization - Fingerprint now uses `git status --porcelain -uall`, so editing a file inside a brand-new untracked directory changes the fingerprint and the "Diff out of date" banner fires (was collapsed to `?? dir/` and hashed as unreadable). - extractTrackedPatchPaths pairs `---`/`+++` and excludes only the file's KEYED path (new side, or old side for a pure deletion) — no longer drops a recreated rename-source file (`git mv a b && touch a`). Verified rm --cached still dedups to one entry and the recreated file still shows. - resolveReviewBase canonicalizes a bare local default name ("main") to its tracking ref ("origin/main") on every call, so a client that loaded before the startup upgrade resolved can't revert the server to the stale local base on the next refresh/switch. Both runtimes (new Pi resolveReviewBase helper). - Removed the dead activeDiffLabel prop chain (FileTree -> DiffTypePicker). - ADR recap: documented the staleness-banner scope limitation (default base only). Verified: typecheck (all projects), bun test 23/23 review-core+fingerprint pass, empirical repro of the untracked-dir fingerprint, rm --cached dedup, and rename-recreate cases. * fix(review): PR-990 review round 4 — dedup content, staging gate, banner timing - Dedup rewrite: instead of dropping the untracked side of a same-path collision (which hid content), drop the tracked DELETION block and keep the untracked working-tree content. `git rm --cached f` + edit now shows the new content, not a phantom deletion — still one entry per path (no dock/nav collision). New stripHeaderPath also strips git's trailing-tab metadata, so space-named files dedup correctly. Verified: rm --cached+modify shows content (1 entry), space-named (1 entry), rename+recreate still shows both. - All-files (and the `a` shortcut on the focused file) now gate staging per-file: committed files in since-base mode are not stageable, matching SectionsPanel / FileTreeNode. Shared isPathStageable helper threaded via ReviewStateContext → AllFilesCodeView. Stops the confusing `git add` no-op that still flipped local staged/viewed state. - /api/diff/switch now awaits recomputeBaseBehindRemote before building the response, so the "Baseline behind GitHub" banner reflects the new base immediately (no ~5s lag switching in, no stale banner switching away). Both runtimes. - ReviewSetupDialog re-applies the recommended default only on first-run dismiss, not when reopened from the header menu (was snapping a mid-session diff back). - SectionsPanel "N added" header now counts sidecar-staged files too, so it matches the staged dots on rows. Verified: typecheck (all projects), bun test 1830 pass, empirical dedup repro, both bundles build. * refactor(review): reuse parsePatchPathToken in removeTrackedDeletions (self-review) Drop the duplicated stripHeaderPath helper — the shared diff-paths parsePatchPathToken already strips a/|b/ prefix + C-quoting + git trailing-tab, and verifies the prefix instead of blindly slicing two chars. Documents the binary-deletion edge (no --- line, so a binary rm --cached is not deduped). * fix(review): PR-990 review round 5 — mixed-base + staging/sort/flicker nits - Atomic base upgrade: the startup origin/* canonicalization swapped currentBase without rebuilding the patch, so /api/diff could advertise origin/main while the served hunks came from local main (mixed-base review on origin/HEAD-absent repos). Now rebuilds the diff for the new base and commits base+patch+ref+ fingerprint together (only if no user switch happened); the fingerprint change makes the client's freshness poll pick it up. Both runtimes. - isPathStageable: gate staging OFF when since-base is active but the sidecar hasn't loaded (was falling through to true, allowing a git-add no-op on a committed file). - SectionsPanel: session-staged files now float to the top of Changes — the sort key was `false ?? stagedFiles.has(...)` which short-circuits; now ORs them. - /api/diff/fresh: early returns now carry baseBehindRemote, so a snapshot change mid-probe no longer clears the "behind GitHub" banner for one poll. Both runtimes. - Removed the now-dead activeLabelFallback prop from DiffTypePicker (FileTree stopped forwarding it in round 3). Skipped (agreed): the fetch-base input-validation nit (no realistic attacker) and the diff-switch TOCTOU (concurrent switches aren't UI-reachable). Verified: typecheck (all projects), bun test 1830 pass, both bundles build. * refactor(review): drop redundant staged-OR in SectionsRow (self-review) item.staged (the grouping key) already ORs in stagedFiles as of the round-5 sort fix, so the row-level `|| stagedFiles.has(...)` is dead weight. * fix(review): surface the startup base upgrade to already-loaded clients Review round 6 fixes: - The startup main -> origin/main upgrade re-baselined the freshness fingerprint, so a client that fetched /api/diff before the rebuild kept the old patch and every /api/diff/fresh probe reported fresh (the probe compares server state to itself, never to what the client renders). Now the fingerprint is only re-baselined when no client has loaded the pre-upgrade snapshot; otherwise the stale baseline trips the normal "Diff out of date - Refresh" banner. Bun + Pi. - preserveFile refreshes (staleness Refresh, post-Fetch) now adopt the server's returned base — exactly the paths where the server may have canonicalized main -> origin/main; keeping the old name sent /api/file-content and Ask AI context against the wrong base. - ReviewSetupDialog: clamp the fixed 800px height to the viewport so the dialog fits on small laptop screens. - Docs: /api/diff/fresh response also carries baseBehindRemote/agentCwd. - Documented the committed-deletion + untracked-recreation dedupe edge as accepted (code comment + ADR) — fixing it needs two same-path diff entries, which the path-keyed UI cannot represent. * fix(review): stop the Git-status default from silently reverting to tree Users with reviewPanelView=sections could keep opening in the tree view on a stale diff type despite their cookie saying Git status. Three causes: - The header Sections toggle persisted the view but not defaultDiffType, creating a conflicted pair (sections + non-since-base default) that every UI writer is supposed to prevent. It now couples the diff default like the setup dialog and Settings do. - configStore's debounced POST /api/config could be lost when a session closed within 300ms of a change, leaving cookie and config.json split. Pending writes now flush on pagehide with a keepalive fetch. - configStore.init() applies config.json over the cookie without the UI coupling, so a stale server value re-corrupted the pair on every load. The app now self-heals at load: if the view says sections but the diff default isn't since-base, it repairs the default (cookie + config.json) and switches the live session to since-base. * fix(ui): keep the stage (+) button border visible on the active file row The button's --border border has no contrast against the active row's 30% primary tint, so it vanished on the selected row until hovered. Tint the border with the row's primary color on active rows (both sections and tree views); the button's own hover border still applies. * ui(review): spell out the panel view toggle — 'Git status | Tree' text instead of icons * fix(review): PR-990 review round 7 — export label, diff snapshot race, unicode paths - exportFeedback describeDiff(): add the missing since-base case — every feedback export in the new default mode read "**Diff:** since-base". - /api/diff GET: snapshot patch/base/ref/error BEFORE the sections-sidecar await and pass the pinned base into buildSectionsSidecar. The startup base upgrade landing mid-await could pair a rebuilt patch with sections grouped against the old base, with initialDiffServed still false so no refresh banner ever came. Bun + Pi. - SectionsPanel: a session-staged untracked file now moves to the Changes section immediately (anticipating the server's next sidecar) instead of sitting in Untracked with a staged dot until refresh. - unquoteGitPath: real C-style unquoting with octal (UTF-8 byte) escape decoding. JSON.parse rejects octal escapes, so non-ASCII names kept their literal \303\251 form — an untracked "café.txt" was silently ABSENT from the review (git diff --no-index could not access the quoted name) and the deletion dedupe Set lookup could never match. getUntrackedFileDiffs now unquotes ls-files output. Unit tests: octal decoding + end-to-end unicode untracked file in a real repo. - review-core.test: worktree subtype round-trip now covers since-base and all (was 6 of 8). Parked (deliberate): SectionsPanel/FileTree keyboard-nav dedup refactor. * fix(shared): unquoteGitPath keeps literal unicode intact (self-review) The byte-collector treated literal non-ASCII code units as single bytes, which would mojibake headers synthesized by our own quoteGitPath (JSON.stringify leaves unicode unescaped inside quotes — workspace-mode prefixed headers round-trip through the same parser). Literal chars now append as string code units (surrogate-safe); only octal escapes go through the UTF-8 byte decoder. * fix(review): PR-990 review round 8 — pre-staged toggles, explicit base, AI context race - useGitAdd: session Set replaced with a tri-state override map folded over the sections sidecar; stagedFiles is now the EFFECTIVE staged set (sidecar + session stages - session unstages). Fixes pre-staged files: the first `a` press actually unstages (was a git-add no-op), the sidebar dot clears after a real unstage (was stuck via sidecar OR), and the All-files header agrees with the sidebar on load. Overrides reset whenever a fresh sidecar arrives (switch, preserveFile refresh, PR response) so stale session intent can't fight new porcelain truth. SectionsPanel drops its own sidecar OR; stagedCount = effective size. - Explicit base picks are honored verbatim: the picker sends explicitBase, and the server permanently disables local-name -> origin/* canonicalization once set (the local/remote groups are distinct choices). The behind-GitHub banner also exempts an explicitly-picked local name — Fetch advances origin/*, so the banner would be un-clearable nagging. Bun + Pi. - buildCurrentAiReviewContext(patch, base): GET /api/diff builds Ask AI context from the same served snapshot as the patch — the startup base upgrade could hand Ask AI a different changeset than the screen. Bun + Pi. - recomputeBaseBehindRemote: capture remoteDefaultInfo once — a concurrent refresh nulling it mid-await threw. Bun + Pi. - Revert stray "Status Update / Updated!" edit to adr/0001 (test debris swept into the round-2 commit). - splitPorcelainRename comment corrected (porcelain v1 does NOT quote plain spaces; a name containing " -> " is ambiguous without -z) + ADR notes for the index-only-changes semantics and the rename edge. * fix(review): FileTreeNode uses the effective staged set — round 9 Round 8 made stagedFiles the effective set (sidecar + session overrides) and removed SectionsPanel's sidecar OR, but missed the same OR in FileTreeNode's since-base row (sectionStaged). In the Tree fallback a pre-staged file unstaged this session kept its dot and the next toggle re-staged it. Grep-swept: this was the last surviving sidecar-staged OR. * refactor(review): make the staged-display invariant unrepresentable The round-8/9 bug class (sidecar staged flag ORed over the effective set) existed because surfaces had a second staging source to reach for. Remove it: stagedFiles is now a REQUIRED prop on SectionsPanel/FileTree/ FileTreeNode and the optional-prop fallback branches reading sectionEntry.staged for display are deleted — a future surface cannot reintroduce the OR because the pattern no longer exists to copy. The sidecar type's staged field and AGENTS.md now document the invariant at the point of temptation. Grep for display reads of .staged now hits only useGitAdd (owner) and the sidecar builder (producer). * refactor(review): self-review cleanups on the staged-invariant hardening - orderFilesBySections: document why its snapshot .staged read is safe (only called at sidecar-fresh moments where snapshot = effective) and that it must not be reused mid-session — the one remaining display-side snapshot read the invariant sweep surfaced. - FileTreeNode: drop the now-pure sectionStaged alias; use isStaged. * fix(review): PR-990 review round 10 — header staging gate, fingerprint cap, single-writer coupling - Single-file diff header now uses the per-path staging gate (canStagePath), closing the last ungated staging trigger: committed-only files in since-base offered a no-op Git Add that flipped local staged/viewed state. Full trigger inventory swept: App `a` shortcut, all-files `a` + header, SectionsPanel rows, FileTreeNode rows, single-file header — all six now per-path gated or group-gated; no context-menu staging exists. - Fingerprint circuit-breaker: the freshness poll's `git status --porcelain -uall` degrades permanently (per cwd, per process) to collapsed -unormal once its output exceeds 2MB — a forgotten node_modules/ no longer burns CPU every 5s for the whole session. Costs untracked-dir edit sensitivity only on such repos; one possibly-spurious staleness banner at the switch. - The sections ⟺ since-base coupling now has a single writer: setReviewPanelView/setReviewDefaultDiffType in @plannotator/ui/config. All five call sites (setup dialog x2, Settings x2, header toggle, first-run reset, self-heal) converted; grep for direct writes of either setting now hits only reviewView.ts. - /api/diff/switch responses pass snapshot args to the AI context builder (both branches, Bun + Pi) — correct today, now robust against future awaits between the epoch check and the response. - Docs: explicitBase in the /api/diff/switch body. Parked per discussion: SectionsPanel/FileTree nav+search+footer dedup (extract when the commit-list view adds a third panel), querySelector row measurement, fetch-base stderr passthrough (standing decision), Pi hasAgentLocalAccess (pre-existing follow-up list). * fix(review): PR-990 review round 11 — subdirectory launches, escape decode, settings note - Repo-root-relative patch paths now resolve against the git toplevel everywhere they meet a filesystem path or pathspec, via a shared resolveRepoToplevel helper: file-content working-tree reads (hunk expansion returned null from a subdirectory launch) and gitAddFile/ gitResetFile (stage/unstage failed with pathspec errors). Both bugs pre-existed for uncommitted/unstaged; since-base made them the default experience. The two existing inline toplevel resolutions (untracked diffs, fingerprint) now use the same helper. Shared code — Pi inherits via vendoring. Real-repo subdirectory tests for both. - unquoteGitPath decodes \uXXXX (JSON.stringify emits it for control chars without a short escape; our synthesized workspace headers round-trip through this decoder). Malformed \u stays literal. Tests. - Pi explicitBase guard matches Bun byte-for-byte on empty-string base. - Settings Git tab notes when the CURRENT repo can't serve the Git-status view (base ref unresolvable) instead of letting the preference look silently broken — it's a global preference, so the options stay. Same-class items left parked (pre-existing, untouched by this PR): open-in root resolution and code-nav file reads from subdirectory launches — on the follow-up list with the Pi divergences. * fix(review): PR-990 review round 12 — per-client freshness, keyboard-operable row controls - Freshness is now judged PER CLIENT: every patch-carrying response (/api/diff, /api/diff/switch, pr-switch, pr-diff-scope) includes snapshotId (the server's draftKey), and the client echoes it on /api/diff/fresh probes. A mismatch reports stale for THAT client regardless of the VCS fingerprint. This fixes the round-12 finding — reloads/second tabs after the startup base upgrade got a permanently bogus staleness banner from the shared pre-upgrade baseline — and deletes the round-6 conditional re-baseline hack entirely (the fingerprint recaptures unconditionally again; the old client's banner now comes from its snapshot mismatch, not a deliberately stale baseline). Also gives unfingerprintable modes (P4, PR layer) snapshot- level staleness for free. Bun + Pi + client hook. - StageControl and ViewedControl (which had the identical gap) are keyboard-operable: tabIndex + Enter/Space activation + focus outline. They're spans inside the row <button> (nested real buttons are invalid HTML), so they need their own focus stop; the a/v shortcuts remain the power path. - Docs: snapshotId on /api/diff, ?snapshot= on /api/diff/fresh. * fix(review): PR-990 review round 13 — re-key snapshots on scope switch, PR-tab refresh - The PR scope switch now re-keys draftKey (= snapshotId + draft storage key) at BOTH commit points, unconditionally — matching every other snapshot commit site. The full-stack branch previously kept the layer patch's key: stale layer tabs never got the banner after a cross-tab scope switch (the exact case snapshotId exists for), and full-stack drafts collided with layer drafts (pre-existing). The layer branch's !layerPatchIncomplete conditional is gone — it only stayed consistent because full-stack never re-keyed. Invariant now: every currentPatch commit is followed by a re-key. Bun + Pi. - Refresh works for any stale PR tab: re-selects the CURRENT scope instead of no-opping for layer (only full-stack could go stale in the fingerprint-only world; snapshot mismatch changed that). Accepted residual (documented in code): after a cross-tab PR switch, refresh updates the patch but not prMetadata — the scope endpoint doesn't carry it, and the full-rehydrate refactor isn't worth the two-tab edge. * fix(review): stale incomplete-layer Refresh uses the non-blocking upgrade path (self-review) Round 13 made Refresh re-select the current scope for stale PR tabs; for an INCOMPLETE layer patch that POST triggers the server's local recompute, which can park for minutes behind checkout warmup — and handlePRDiffScopeSelect renders the full-screen switch overlay the whole time. Route that case through handleLoadFullDiff (same POST, progress notice instead of modal), which exists for exactly this slow path. * fix(review): round 14 — composite snapshot id, fully-pinned /api/diff, "All changes" label - snapshotId is now content hash + diff type (+ PR scope), built by a single currentSnapshotId() helper used at every response site and the freshness compare. A cross-tab MODE switch with a byte-identical patch (layer vs full-stack on a single-PR stack) now flags old tabs; the base is deliberately excluded so a same-commit main -> origin/main canonicalization stays banner-silent (round-12 noise-avoidance kept). draftKey stays a pure content hash — drafts survive content-identical round-trips. Bun + Pi. - GET /api/diff pins ALL served fields (diffType, hideWhitespace, prDiffScope join patch/base/ref/error/snapshotId) and the sidecar + AI context builders take the pinned type instead of reading globals — a concurrent tab switch during the sidecar await can no longer produce a since-base patch labeled with another mode. Bun + Pi. - First-user feedback: the since-base label is now plain English — "All changes since origin/main" (dynamic, follows the picked base); dialog/Settings short form "All changes"; feedback exports match. uncommitted reverts to "Uncommitted" where it had borrowed "All changes", so the two stay distinguishable side by side. * docs(review): finish the Since-main -> All-changes terminology sweep in comments (self-review) * fix(review): PR-990 review round 15 — base-revert race, freshness reset, unstage regroup - resolveReviewBase gains a second, probe-independent rule: a non-explicit echo of the bare local name of the CURRENT origin/* base stays on the tracking ref. The existing rule keys off remoteDefaultInfo, which comes from a second network probe that can lag the startup upgrade by seconds; in that window a diff-type/whitespace switch echoing "main" committed the session back onto the stale local branch and set baseEverSwitched, permanently blocking the upgrade. Bun + Pi. - useDiffFreshness resets stale/dismissed state on snapshotId too — since round 14 a new snapshot can reuse identical patch text with a different id, and the old banner state wrongly carried over until the next poll. - SectionsPanel: unstaging a PRE-staged add moves it back to Untracked (mirror of the round-7 stage regroup). Detection deliberately uses the sidecar's snapshot flag to recognize "was pre-staged"; staged modifications stay in Changes, staged renames remain a refresh-heals edge. * fix(review): PR-990 review round 16 — rename staged count, guarded fetch-base replay - "N added" no longer double-counts staged renames: the sidecar truthfully marks BOTH porcelain sides staged, but an above-threshold rename renders as ONE file — the hidden old path inflated the count and left a phantom effective-staged entry that unstaging the visible row couldn't clear. The client now filters sidecar-staged paths to rendered files, which is correct in both patch shapes (below-threshold renames render delete+add as two rows and both sides pass the filter). Client-only; the sidecar stays faithful to porcelain. - The fetch-base completion only replays the diff refresh if the user's diff type/base selection is unchanged since the click — a slow fetch no longer yanks the review back to the view captured at click time. Declined with reasons (in review thread): first-run persist-before-confirm (approved forced default, second flagging), explicit-base flag ordering (early-set preserves user intent; commit-after-win would leave picker and server disagreeing), base-picker revert target (near-unreachable compound race, deferred). * fix(review): PR-990 review round 17 — find origin/main on feature-only clones getDefaultBranch's chain (origin/HEAD -> local main -> blind "master") skipped the fetched remote-tracking ref entirely. On checkouts with no origin/HEAD symref and no local main/master — CI checkouts, `clone --branch feature`, extra worktrees — it guessed "master", the base didn't resolve, getGitContext suppressed since-base, and the flagship "All changes" view silently disabled itself for the whole session even though origin/main was fetched and diffable. The startup upgrade can't rescue it either (its guard reads the bogus "master" as a deliberate non-default base). Chain is now: origin/HEAD (verified) -> origin/main -> local main -> origin/master -> "master" — remote-tracking refs preferred, matching the function's stated prefer-upstream intent. Shared core (Pi inherits); real-repo test reproducing the exact clone shape.	2026-07-04 09:53:49 -07:00
Michael Ramos	cb6667e991	fix(ai): drive Codex Ask AI via codex app-server (#971) ... * fix(ai): drive Codex Ask AI via app-server + answer-first review prompts (#971) Codex Ask AI previously ran via @openai/codex-sdk (codex exec), which forces approval_policy=never and breaks in enterprise-managed Codex environments that ban it (#971). Replace the transport with a long-lived 'codex app-server' process over JSON-RPC. - New provider packages/ai/providers/codex-app-server.ts (registered as 'codex-sdk' to preserve cookie/agents.ts/UI-gate); omits approvalPolicy so Codex resolves the user's + managed policy, pins read-only sandbox, and surfaces interactive approvals through the existing PermissionCard. - Delete codex-sdk.ts and drop the @openai/codex-sdk dependency (and its 6 prebuilt platform binaries); gate registration on 'which codex'. - SessionManager: additive, optional dispose?() hook to kill the spawned process on evict/remove — a no-op for Claude/OpenCode/Pi (they don't implement it). Also rework the Ask AI prompts (all providers, separate from the transport): - Every mode now instructs the agent to answer the user's message directly and not launch an unprompted review of the context. - Code review stops pasting the whole diff for git-reproducible diff types and instead tells the agent how to inspect it (git diff <base>..HEAD, three-dot for merge-base); non-git/PR/workspace types still paste. - Claude gains the Bash tool so it can run git (still gated by approvals). - The UI passes diffType/base (session) and what the user is viewing (per question) into the context. Verified: full typecheck, full test suite (101 ai tests), and a live end-to-end smoke against codex app-server. * fix(review): pin AI approval card above the input/model bar Render pending approval cards just above the input + provider/model bar in both the document chat (DocumentAIChatPanel) and code-review AI tab, instead of at the top of the scroll, so the user sees them where they act. * fix(ai): harden Codex abort/cancel + add Ask AI Stop button Addresses code-review findings on the codex app-server provider: - turn/interrupt was sent as a notification (no id) so Codex ignored it and abort never took effect. It's now a proper JSON-RPC request. - Filter turn events/approvals by turnId and reject an aborted turn's stragglers, so a stopped turn can no longer leak output into — or prematurely finish — the next turn (ask-stop-ask race). - Guard listeners by query generation and end the drain loop on the abort signal, so a superseded/stopped turn can't touch the live one and abort returns promptly instead of waiting for turn/completed. - Handle abort during startup: once the turn id is known, interrupt it instead of running it in the background. - Add a sendAndWait timeout so a stalled (alive-but-unresponsive) process errors instead of hanging forever. - Drain stderr (stdio 'ignore') to avoid a pipe-buffer deadlock. Also add a Stop button to both Ask AI surfaces (plan/annotate DocumentAIChatPanel and code-review AITab via ReviewSidebar). It replaces Send while streaming and calls the hook's abort -> /api/ai/abort -> session.abort(); the hook already exposed abort but nothing surfaced it. * feat(ai): drive Codex models + reasoning levels from model/list - Codex provider fetches model/list at startup (throwaway app-server, like Pi/OpenCode) and populates the real models plus each model's actual supportedReasoningEfforts + defaultReasoningEffort. Replaces the hardcoded model list and the static AI_REASONING_EFFORTS (which mislabeled xhigh as 'Max' and omitted minimal). - AIProviderBar + AIConfigBar now show the selected model's real efforts and hide the control when a model reports none. xhigh is shown verbatim. - Fix: the Stop button now also appears in the populated code-review chat state (a prior edit missed the second GeneralInput due to indentation). * fix(ai): scope Claude Ask AI Bash to read-only git; clear stale approval cards on Stop - Claude Ask AI no longer auto-allows bare Bash (which ran arbitrary shell with no Allow/Deny prompt). Replace it with scoped read-only git rules (Bash(git diff:*), show, log, status, rev-parse, merge-base, ls-files). git reads auto-run so the agent can inspect large diffs itself; any other command (write git, arbitrary, or injected compound) falls through to the permission card. Keeps the git-inspect approach (large diffs don't fit in the prompt) while closing the auto-exec hole. - useAIChat.abort() now drops still-undecided permission cards: abort cancels them server-side, so leaving them visible was a dead Allow/Deny. * feat(ai): code-review Ask AI shares the agent-review prompt machine, delivered as user messages Code-review Ask AI built its own diff description (gitInspectInstruction) in the system prompt from just diffType+base, which was wrong for full-stack, hide-whitespace, untracked files, and PR worktrees. Replace it with the same machine the review jobs use, delivered on the user's messages. - Server: buildCurrentAiReviewContext() reuses buildAgentReviewUserMessageForTarget (contextOnly) for the current view and ships it as aiReviewContext in every diff payload (/api/diff + switch/PR/scope). Mirrored in the Pi server. - Client: review-editor latches aiReviewContext onto each question via the pure buildReviewContextPreamble (packages/ui/utils/aiPrompt.ts) and buildDefaultPrompt — full block on the first message / when the view changes (incl. after a provider switch via the !sessionId fresh-session check), a short reminder otherwise (never re-pastes a large diff). - context.ts: delete the duplicate gitInspectInstruction; code-review system prompt is now role-only. Providers untouched (provider-agnostic user message). - Tests: machine scenario gaps (plain PR, full-stack default, PR-worktree origin/<base> + stale-main warning, untracked mention, jj-evolog, workspace lines); composition (first/reminder, command/pasted, preamble ordering). * fix(ai): agentic remote-PR context, UTF-8 stream decode, real Stop on supersede/disconnect Addresses review findings on the Ask AI prompt work: - Remote PR without a confirmed local checkout no longer gets URL-only. The agent is told it's in a PR worktree that's being prepared, to verify the PR files exist before relying on them, and to diff with git diff origin/<base>... HEAD (URL fallback). Inform + trust the agent rather than pasting. Shared machine, so review jobs get the same framing. - Decode Codex stdout with a streaming TextDecoder instead of per-chunk toString(), so multi-byte UTF-8 split across chunks no longer corrupts into U+FFFD (matches the Pi provider). - Stop now actually stops the server turn: ask() awaits /api/ai/abort when a new question supersedes a streaming one (awaiting avoids racing the new query into session_busy), and the /api/ai/query SSE stream gains a cancel handler so tab-close/navigation aborts the turn too. Both reuse the existing per-provider session.abort(); factored a shared postServerAbort helper. * fix(ai): per-model reasoning effort, shared PR-checkout readiness, Stop-then-ask race, type hole - Reasoning effort is now tracked per model (a map keyed by model) instead of one global value, so switching to a model that doesn't support the prior level (e.g. xhigh) no longer posts a stale/unsupported effort that Codex rejects. Each model keeps its own level; nothing leaks across. - Extract resolvePoolCwd into packages/shared/worktree-pool.ts (ready/pending/ absent) and use it from both servers' resolvePRLocalCwd so the readiness rule can't drift. Fix the Pi Ask AI helper to ready-check like Bun, so a warming PR checkout no longer claims 'checked out at PR head' and misdirects the diff. - Stop-then-ask no longer races into session_busy: the abort promise (from Stop or a superseding question) is stashed and the next ask() awaits it before sending. Stop still kills the turn instantly; a follow-up just waits the one abort round-trip. - Declare aiReviewContext on the initial /api/diff response type. * fix(ai): surface real Codex error messages, skip transient retries, auto-deny permission escalations Validated against codex-rs: - Codex's ErrorNotification nests the text at params.error.message (TurnError.message); we read params.message → always 'Unknown error'. Read the nested field (top-level fallback) so auth/usage-limit/stream failures show their real, actionable text. - Skip transient error notifications (willRetry=true): Codex retries on its own and the turn continues, so surfacing them flashed a spurious failure before the real answer. - Handle item/permissions/requestApproval: respond {permissions:{}, scope:'turn'} — byte-for-byte Codex's own cancel response (codex_delegate.rs) — instead of 'Unsupported request'. Ask AI is read-only, so denying the escalation lets the turn continue sandboxed rather than failing. Interactive grant deferred. * refactor(ai): share AI provider/model config in one hook; aggregate paginated model/list - Extract useAIProviderConfig (packages/ui/hooks): one home for provider/model/ reasoning-effort selection — initial state, auto-resolve on capabilities load, per-model effort (no leak across models), and persistence. Both the plan and code-review apps now call it and only compose the session reset (the hook can't own reset without a cycle through useAIChat). Plan editor adapted to the effect-based resolve (adds aiDefaultProvider state). This fixes the plan editor's stale-effort-on-model-switch bug by construction and stops the two apps' copies from drifting again. - fetchModels now follows model/list's nextCursor and aggregates every page into one list, so larger model catalogs aren't silently truncated (with a page guard against a misbehaving cursor). Skipped per discussion: legacy v1 approval methods (we're a v2 client). * feat(agents): per-agent review default; bound Codex model discovery so it can't stall the AI panel - The selected review profile is now tracked per review engine (claude/codex/ cursor/opencode) instead of one flat global value, so each agent keeps its own review default. Public hook API (reviewProfileId/setReviewProfileId) is unchanged — getter derives the current engine's value, setter writes it — so AgentsTab needs no changes. One-shot migration seeds every engine with any existing flat pick. Adds parseReviewProfileByEngine + tests. - fetchModels (Codex model discovery) now uses a short 6s timeout for its initialize + model/list RPCs instead of the 30s default. /api/ai/capabilities awaits model discovery, so an installed-but-unauthenticated codex could otherwise block the AI panel for ~30s; it now falls back to the static model list fast. Authed codex is unaffected (discovery completes in tens of ms).	2026-06-28 09:58:09 -07:00
Michael Ramos	e138c824aa	feat(review): Cursor + OpenCode review engines (unified marker-review) (#959) ... Add Cursor CLI and OpenCode as background code-review engines alongside Claude/Codex/Tour, unified behind a shared marker-review module (prose marker-block parsing, since neither CLI exposes a schema flag). Includes multi-provider dynamic model discovery, review-profile support, fail-closed parse/ingest, Bun+Pi parity, and the review UI (Provider picker, brand icons). Hardening from review: per-job nonce on marker tags (prevents echoed/quoted tags from corrupting extraction), truncation fail-closed, async model discovery (never blocks the event loop), nested-slash model ids, shared finding types.	2026-06-24 14:35:52 -07:00
Michael Ramos	affeaa07fc	feat(review): custom reviews as Agent Skills + whole-file/general findings (#955) ... Custom reviews: pick an Agent Skill and its SKILL.md body becomes the review prompt (read live from global skill folders, never copied; default review byte-identical). Agent findings can be a line, a whole file, or a general review-level comment, and are never silently dropped. Plus: agent-icon engine picker, icon-only file-header opener, semantic diff restored as its own dock panel, and a dead-code sweep.	2026-06-23 20:49:48 -07:00
Michael Ramos	740d6fb2eb	Add WebTUI agent panel to annotate mode (#941) ... * feat(annotate): add WebTUI agent terminal * feat(annotate): wire WebTUI agent into annotate UI * docs: recap annotate agent terminal work * fix(annotate): harden agent terminal runtime * docs: add annotate agent terminal runtime ADRs * fix(annotate): polish agent terminal integration * fix(ui): preserve comment draft on Ask AI failure * fix(annotate): address terminal review findings * fix(annotate): harden agent terminal runtime fallback	2026-06-19 09:04:15 -07:00
Josemi Liébana	b7ef0756d9	fix(annotate): add /api/save-notes POST endpoint to annotate server (#884) ... * fix(annotate): add /api/save-notes POST endpoint to both servers Copies the save-notes route from the plan review server into the annotate server (Bun source and Pi extension copy), enabling Save to Obsidian in annotation mode. Fixes #844 * test(annotate): add saveToObsidian unit tests and HTTP endpoint tests Verifies saveToObsidian writes files correctly and handles missing vaults. HTTP endpoint tests cover success, empty integrations, and integration-level error (not 500). Imports consolidation from ./integrations into a single statement. * fix(annotate): normalize server port fallback * refactor(server): extract shared handleSaveNotes handler, fix catch-block bug Move the /api/save-notes logic into shared handler modules (shared-handlers.ts for Bun, handlers.ts for Pi) following the existing pattern for handleImage, handleUpload, handleDraftSave. Replaces four inline copies with two canonical implementations. Fixes: - Bun annotate catch block now correctly returns 500 (was logging only) - Misindented brace in Pi serverAnnotate.ts resolved by extraction - Revert unrelated port fallback change (keep server.port! for consistency) - Static imports in integrations.test.ts - Add /api/save-notes to CLAUDE.md Annotate Server API table * fix(opencode): inject annotate server starter instead of global mock.module commands.test.ts mocked the annotate server with `mock.module("@plannotator/server/annotate", ...)`. Bun module mocks are process-global and cannot be unset (oven-sh/bun#7823, #12823), so the stub leaked into every suite that runs after it — in particular any test that boots the real annotate server received a stub with no `.url`. Make `startAnnotateServer` injectable through the existing CommandDeps (defaulting to the real import, so production is unchanged) and have the test pass its stub that way. This keeps the fake local to the opencode suite and unblocks real annotate-server integration tests. * test(server): cover save-notes — handler unit tests + annotate e2e wiring - shared-handlers.test.ts: unit-test handleSaveNotes directly (Obsidian write, empty integrations, integration-error reported not thrown, 500 on bad body). - annotate.test.ts: boot the real annotate server and POST /api/save-notes, asserting it is served as JSON (not the SPA HTML catch-all) — the regression guard for #844. Now possible because the opencode suite no longer installs a global annotate module mock. --------- Co-authored-by: Michael Ramos <mdramos8@gmail.com>	2026-06-16 21:38:39 -07:00
Yoni Hornstein	201ca11ec4	feat(config): support share toggle via config.json (#921) ... Allow disabling URL sharing through ~/.plannotator/config.json ({ "share": "disabled" }) in addition to the PLANNOTATOR_SHARE env var. Adds a resolveSharingEnabled() helper (env var > config > default enabled) and routes all sharing checks through it across the hook server, OpenCode plugin, and Pi extension. Docs updated.	2026-06-16 18:47:24 -07:00
Michael Ramos	be2d06a7c2	Make HTML annotations render HTML by default ... * feat(annotate): render html files by default * fix(annotate): support raw html assets and sharing * fix(annotate): address html first review followups * fix(editor): avoid raw html sidebar init crash * fix(annotate): support portable html shares * fix(annotate): harden html share support * fix(share): clear attachments when loading shared payloads * fix(share): warn on remote share link failures * perf(annotate): lazy-build html share payloads * test(annotate): guard lazy html share generation * test(annotate): drop flaky html share server test	2026-06-16 16:16:05 -07:00
Michael Ramos	d3fd62297b	feat(readme): redesign README (#891) ... * feat(readme): redesign README — hero banner, agent icons, simplified install - New centered hero: banner, tagline, theme-aware agent icon row (picture/prefers-color-scheme), demo links - Intro paragraph naming all nine supported agents - Feature cards with screenshots; dedicated Annotate HTML Artifacts, Code review, and Sharing sections - Install collapsed to one shared installer + per-agent table covering all nine agents, linking each app README - New Try it section for manual invocation after install - Fixed Codex plan-mode claim (supported via Stop hook), completed PLANNOTATOR_ORIGIN values, fixed demo links and typos - Added Integrations, Remote/SSH, Security, and Configuration sections * feat(readme): consolidate command examples into a Commands section - Annotate HTML Artifacts is now screenshot-only - Drop the duplicate code review screenshot (code-review-thumbnail.png removed) - New Commands section: Annotate, Code review, and CLI groups * fix(readme): point doc-review demo at Pi video, drop wrong code review demo link * style(readme): plain-language pass - No em dashes anywhere; split into sentences or swapped for colons/commas - Rewrote the intro: integration with the agent session is now explicit (plugs into hooks/commands, feedback lands back in the live session) - Varied repeated phrasing between the two feature cards - Integration labels use colons instead of dashes * style(readme): simplify feedback phrasing in hero and feature card * feat(readme): add Plan mode note and sessions command to Commands section * feat(readme): sharing beta link and Workspaces signup CTA badge * style(readme): drop divider between feature table and HTML artifacts section * style(readme): bump banner width 512 -> 640 * style(readme): fix agent icon sizing - Crop Pi icon viewBox to glyph bounds (was 40% padding) - Render Amp at 32px to offset its thin mark * style(readme): bump Amp icon to 38px for visual parity * fix(readme): strip em-based size attrs from Amp icon so it scales The svg root had width/height of 1em plus an inline style, the only icon in the set with relative sizing. SVG-as-image resolves em at a fixed 16px and the drawing stayed pinned there regardless of the img height. Removed the attrs, back to the uniform 28px row height. * style(readme): width=100% on feature table images * style(readme): cmux-style 40/60 feature table with middle-aligned text * style(readme): split feature rows into separate tables so zigzag keeps 40/60 * style(readme): single feature table, text left 40 / image right 60, no zigzag * style(readme): tighten intro and feature card copy * style(readme): remove stray space before period * style(readme): another round of copy edits * chore: remove internal docs notes, move readme-assets to .github/assets - Deleted docs/ (three internal handoff notes, nothing referenced them) - README media now lives in .github/assets/ instead of a root-level folder - All 18 asset links in README updated * style(readme): split intro into two paragraphs * feat(readme): live collaboration teaser in Sharing, tighten encryption copy * style(readme): retitle Sharing & Multiplayer, rework CTA badge to banner palette with balloon * feat(readme): updated annotate and review screenshots * feat(readme): updated HTML artifact screenshot * feat(readme): swap HTML artifact screenshot for v2 * feat(readme): restore share workflow sentence and Codex command note * docs(agents): document bun link workflow for local plannotator command * style(readme): trim Windows note from Codex command tip * feat(readme): one-line AI features note under feature table	2026-06-11 13:10:50 -07:00
Michael Ramos	afac44b940	feat(review): migrate all-files code review to Pierre CodeView (#885) ... Replaces the all-files renderer with Pierre's virtualized CodeView (one scroller, identical look, scales to huge diffs); deletes the legacy AllFilesDiffView/LazyFileDiff and their flag; single-file panels stay on FileDiff. @pierre/diffs pinned exact 1.2.8 (1.2.9's tree is broken on npm — see #880). Also ships: diff staleness detection ('Diff out of date · Refresh' toolbar notice backed by shared per-VCS fingerprints + GET /api/diff/fresh on both Bun and Pi servers), a stale-content guard so files edited mid-review degrade to raw-patch view instead of breaking virtualization, the Ask AI wrong-file fix (pre-existing), and toolbar/header UI polish (global settings cog, aligned sem badges, no all-files split dragger). Hardened through five review waves; tested with 1384 passing tests incl. real-git fingerprint and patch-consistency suites. Known minor gaps documented in the PR.	2026-06-10 21:07:58 -07:00
Michael Ramos	c3ba55e889	feat(review): add semantic diff overview (#871) ... * feat(review): add semantic diff overview * fix(review): address semantic diff review findings * fix(review): harden semantic diff fallback * fix(review): avoid repo-local sem execution * fix(review): normalize semantic diff cwd * test(review): cover semantic diff local cwd * fix(gitlab): fetch raw MR diffs * fix(review): harden sem path resolution * docs(review): add semantic diff handoff * feat(review): semantic diff cards, header sem badges, jump-to-line - Restyle the semantic panel as real bordered cards (shadcn surface) instead of ASCII box-drawing; centered column, grid-aligned entity rows. - Hide orphan (module-level) changes from the rows, matching sem's default; the count still surfaces in the summary line. - Add a 'sem · N' hover popover to each diff file header showing that file's semantic changes, reusing the same row component as the panel. - Clicking an entity (panel or popover) scrolls the diff to the lines via pierre's [data-selected-line]; centers only when off-screen so manual drag-selection isn't disturbed. - Extract shared rows/helpers into semanticDiffShared; share one cached /api/semantic-diff fetch across header badges. * feat(review): land on All files by default Semantic diff stays available via the file-tree nav entry and header badges, but it's no longer the initial landing view. * chore(review): drop dead change-symbol entries, log badge fetch failures - Remove unreachable 'moved'/'renamed' entries from the changeSymbols table (getChangeSymbol early-returns for those before the lookup). - Log a console.error once per patch when the file-header badge's semantic diff fetch fails or returns a non-ok status, so a systemic failure leaves a trace instead of every badge silently showing nothing. * feat(review): flatten semantic diff panel into grouped list - Dissolve the per-file bordered cards into flat sections: the file path is now a quiet underlined header (single hairline) with entity rows flush beneath, hierarchy from whitespace + hover tint instead of boxes. - Split the path into muted directory + emphasized filename for faster scanning. - Nudge the add/remove glyphs (⊕/⊖) ~15% larger, line-height pinned so rows don't grow; modified/rename/reorder glyphs unchanged.	2026-06-08 22:23:54 -07:00
Michael Ramos	b19505efd3	chore: remove the redundant /plannotator-status and /plannotator-archive commands (#873) ... Two agent command-surface cleanups. Both remove only the command entry points; all underlying infrastructure stays. 1. /plannotator-status (Pi): removed — it echoed phase/plan-file/progress on demand, but that state is already shown ambiently (status bar + live checklist widget). The phase/checklist state machine is untouched. 2. /plannotator-archive (all agents): removed the command/skill entry points across every surface — Claude/Codex/Kiro skills, Pi, OpenCode (handler + dispatch + cli-bridge + embedded + stub), Droid, the Kiro agent prompt, all three installers, docs, marketing, and the CI deprecated-command guard. The installers also gained a stale-skill cleanup so upgraders drop a previously installed plannotator-archive skill. Kept (infrastructure) — archive browsing stays available in-review via the sidebar: the `plannotator archive` CLI subcommand (apps/hook/server), the mode:"archive" server path + /api/archive endpoints, ArchiveBrowser/useArchive, the sidebar Archive tab, sessions.ts "archive" mode, and ~/.plannotator/plans storage. Verified: bun test scripts/install.test.ts → 72 pass; pi-extension typecheck + build:opencode pass; repo-wide residual scan clean; KEEP-set integrity confirmed; one orphaned import (opencode commands.ts) caught in self-review and removed.	2026-06-08 11:08:11 -07:00
Oscar Silva	f08764063d	feat(review): support multi-repo workspace reviews (#543) ... * feat(review): support multi-repo workspace reviews (#527) * fix(workspace): address critical issues from deep review - Fix race condition in label generation by pre-computing labels sequentially - Fix rewritePatchLine to support quoted paths and rename/copy headers - Add separator between aggregated patches to avoid invalid diffs - Normalize input paths in resolveWorkspaceFilePath - Add timeout to PR discovery (15s) to prevent server hangs - Fix PATCH /api/workspace/repo to rollback state on failure via applyRepoMutation - Validate body.source runtime (must be 'local' or 'pr') - Snapshot active repo in agent jobs at launch to prevent race in onJobComplete - Prevent double-prefixing of agent findings when paths are already prefixed - Fix frontend findWorkspaceRepoForPath to use longest-prefix matching - Fix shared types: diffType uses DiffType, platformUser is string | null * fix: remove duplicate gitRuntime export in vcs.ts * fix: resolve remaining merge conflict in local review mode, remove stale detectManagedVcs import * Add local multi-repo workspace review support * Fix workspace review edge cases * Remove session query from browser launch * Add switchable workspace review modes * fix(review): cover opencode workspace bridge * fix(review): clean workspace review plumbing * fix(review): clarify workspace agent finding paths * fix(review): preserve diff paths with spaces --------- Co-authored-by: Michael Ramos <mdramos8@gmail.com>	2026-06-05 14:29:38 -07:00
Michael Ramos	26ca4e0275	Single-source skills (core/extra), replace Claude Code commands with skills, de-hardcode installers (#850) ... * feat: single-source skills into core/extra, replace Claude commands with skills, de-hardcode installers - apps/skills/core/{review,annotate,last,archive}: single authoritative source for the always-installed skills (archive is new); all carry disable-model-invocation + agents/openai.yaml sidecars - apps/skills/extra/{compound,setup-goal,visual-explainer}: no longer default-installed (except Kiro); installers print an `npx skills add backnotprop/plannotator/apps/skills/extra` suggestion - Claude Code: apps/hook/commands/ deleted, command heredocs removed; core skills in ~/.claude/skills are the slash commands now - Installers: OpenCode/Gemini command files copied from an extended sparse checkout instead of heredocs; install.cmd gains the previously missing OpenCode command install; aggressive cleanup of legacy ~/.claude/commands and ~/.codex/skills artifacts - Codex: core skills install to ~/.agents/skills (official path); ~/.codex/skills install removed - Pi: extension no longer bundles skills; #670 settings filter removed * fix: review findings — old-tag soft guards, cmd replace-not-merge, plugin-update hint, frontmatter test - install.sh: a --version tag predating apps/skills/core no longer aborts the whole copy subshell (which also skipped OpenCode/Gemini commands); core skills now soft-skip with an accurate message, matching ps1/cmd - install.sh: subshell failure message no longer claims "git required" when git was present (clone/network errors get their own wording) - install.cmd: pre-remove skill dirs before xcopy so upgrades replace rather than merge (stale files from renamed/deleted skill files no longer linger; parity with sh/ps1) - all installers + docs: tell upgraders to run /plugin marketplace update so the plugin's old namespaced plannotator:* commands disappear (#817) - install.test.ts: assert every core SKILL.md sets disable-model-invocation: true — the load-bearing line that keeps core skills out of Pi's system prompt (#842 regression guard) * test: pin old-tag soft-guard behavior, dedupe core-skill list in tests * fix: interrogation review findings — cross-installer diagnostic parity - install.ps1/install.cmd: emit the "predates the core/extra skill layout" diagnostic on old pinned tags instead of silently skipping core skills (parity with install.sh) - install.ps1: clone/network failure no longer claims "git required" (git was already verified present); the outer catch now reports the actual exception - install.sh: "Installed OpenCode/Gemini commands" echoes are guarded on the copy actually having a source, so old pinned tags don't print false success (ps1/cmd already gated this way) - AGENTS.md: opencode-plugin commands/ comment now reflects all four command stubs - install.test.ts: shared test asserts the soft-skip diagnostic exists in all three installers and pins ps1's honest failure wording * fix: respect CODEX_HOME for Codex home directory (#852) Codex stores config and state under $CODEX_HOME when set, falling back to ~/.codex (developers.openai.com/codex/config-advanced). Plannotator hardcoded ~/.codex in two places: - runtime: codex-session.ts scanned ~/.codex/sessions for rollout files, so `plannotator last` failed with "No rendered assistant message found" when CODEX_HOME pointed elsewhere. Now resolved the same way copilot-session.ts handles COPILOT_HOME and session-log.ts handles CLAUDE_CONFIG_DIR. - installers: detection, config.toml/hooks.json paths, manual-setup instructions, and the stale-skills cleanup now derive from CODEX_HOME in all three scripts. Tests: codex-session.test.ts covers rollout discovery under a CODEX_HOME temp dir; install.test.ts asserts all three installers respect the variable and that the fallback is the only hardcoded ~/.codex path left in install.sh. * fix: hard-fail skill install, guard command cleanup, one-time extras migration External review triage on PR #850 surfaced two real installer issues: P1 — commands deleted before replacement: the Claude command cleanup ran before the git-gated skill install, so a missing git, a failed clone, or an old pinned tag deleted the user's slash commands and installed nothing (a regression — the old installer needed no git). Now: - missing git is a hard failure before anything is touched ("install git, then run this installer again") - a failed fetch is a hard failure ("something went wrong — run the installer again") instead of a silent skip - the legacy command cleanup runs AFTER the install and only removes a command file when its same-name replacement skill exists on disk - old pinned tags keep the soft-skip (no deletion happens, commands survive, CI e2e against old tags stays green) P2 — recurring extras deletion: the extras cleanup ran on every invocation, deleting copies users reinstalled via the suggested `npx skills add` (the copies are byte-identical, so only provenance can tell them apart). The cleanup is now a one-time migration recorded in a migrations ledger under the Plannotator data dir (<PLANNOTATOR_DATA_DIR|~/.plannotator>/migrations/), the same record-what-you-did pattern package managers use. All three installers (sh/ps1/cmd) updated in parity; tests pin the guard condition, the ledger gating, and the hard-fail messages. * test: tripwire — install.cmd must never contain /dev/null redirects * fix: every skill sets disable-model-invocation — no exceptions Maintainer rule: all Plannotator skills are user-invoked, never model-auto-invoked. setup-goal (missing since #665) and the three Kiro skills now carry the flag. The frontmatter test scans every SKILL.md in apps/skills/core, apps/skills/extra, and apps/kiro-cli/skills dynamically — with a floor of 10 — so a future skill cannot ship without it. * docs: git is a hard installer requirement; clarify post-gate sections complete on re-run * docs: align ps1/cmd comments with hard-fail semantics * feat: guided install — extras opt-in via skills CLI, model-invocation picker Interactive terminals get a two-question wizard on first run: 1. Install the extra skills? Yes delegates to `npx skills add backnotprop/plannotator/apps/skills/extra` (its UI picks the agents), wired to /dev/tty so piped curl|bash installs still work. Skipped when extras already exist on disk. 2. Make any skills callable by the model? Yes opens a space-toggle checkbox (sh/ps1) or numbered toggles (cmd), listing all skills if extras were chosen, core-only otherwise. Chosen skills get disable-model-invocation stripped from their INSTALLED copies and the Codex sidecar's allow_implicit_invocation flipped — re-applied every run since installs replace skill folders. Repo sources stay locked. Answers persist to <data dir>/install-prefs (shared format across all three installers) and re-runs reuse them silently; --reconfigure re-opens the wizard. Automation is untouched: no terminal means no prompts and today's defaults; --extras/--no-extras/--model-invocable/ --non-interactive give scripts explicit control. * fix: self-review of guided install — cmd pipe expansion bug, flag/wizard interplay - install.cmd: the checkbox preselection used `echo !var! | findstr` — each side of a cmd pipe runs in a child WITHOUT delayed expansion, so the saved choices passed through as literal !var! text and preselection never matched. Replaced with a substring-replace containment test (no pipe). - all three: a wizard question whose answer was already provided by a CLI flag (--extras/--no-extras/--model-invocable) is no longer asked and then silently overridden — the flag pre-answers it. - install.cmd: unknown-option usage line now lists the wizard flags. * feat: guided install question 3 — install Glimpse (native window) glimpseui (third-party npm package, PR #840) gives Plannotator a native WebView window instead of a browser tab; the runtime already auto-detects it on PATH, so a global install is all that's needed. - Wizard asks "Install Glimpse?" (default yes) after the skills questions; skipped when glimpseui is already on PATH - Yes runs `npm install -g glimpseui` (bun fallback on sh/ps1; printed instruction when neither exists) — wizard or explicit flag only, silent re-runs never install software - --glimpse / --no-glimpse flags for automation; choice persisted to install-prefs like the others - docs + tests updated (glimpse detection, install command, flags, and persist-condition assertions across all three installers) * fix: self-review of Glimpse question — cmd bun fallback, stale usage text * fix: merge-window hardening — guard Codex cleanup, remove old-installer junk dirs plannotator.ai serves install.sh live from main (public/ symlink, deployed on push), while the script fetches repo files at the LATEST RELEASE TAG. Between merging the core/extra restructure and cutting the release that ships it, the live script runs against the old-layout tag. Two hazards in that window: 1. The Codex stale-skill cleanup removed working ~/.codex/skills with no successor installed (core skills soft-skip on old tags). Now the cleanup runs AFTER the install and removes a core skill only once its replacement exists in ~/.agents/skills — same guard the Claude command cleanup uses. The compound/setup-goal stale copies stay unconditional (never Codex's to begin with). 2. The reverse combo (cached OLD script + NEW release tag) wholesale- copies apps/skills/* and leaves junk core/ and extra/ directory copies in ~/.claude/skills. Never valid skill names — all three installers now remove them on every run. * fix: glimpseui is a devDependency — consumers never use it from node_modules PR #840 added glimpseui to dependencies in @plannotator/server and @plannotator/pi-extension, but nothing imports it: both runtimes detect the CLI on PATH (Bun.which / a manual PATH walk) and spawn it. The dep only ever mattered in repo development, where `bun run` prepends node_modules/.bin to PATH. For consumers it was inert download weight — OpenCode plugin installs and `pi install` pulled a third-party package that could never be detected (Pi's loader does not expose node_modules/.bin; verified). Moved to devDependencies in both: dev flows keep working, published packages stop shipping it. The sanctioned end-user path is the guided installer's global `npm install -g glimpseui`. * fix: clean stale plugin command files from the installed plugin checkout (#817) The installer already manages hooks.json inside ~/.claude/plugins/marketplaces/plannotator/apps/hook/, so the earlier "don't reach into plugin storage" rationale for leaving the old namespaced plannotator:* command files there was inconsistent. All three installers now remove them — same replacement-skill guard as the bare ~/.claude/commands cleanup — making the #817 duplicate menu entries die on a single installer run + restart instead of waiting for /plugin marketplace update. Hints/docs reworded accordingly. * Revert "fix: clean stale plugin command files from the installed plugin checkout (#817)" This reverts commit 3df37da92c. * fix: Windows CI — assert the new no-commands contract; gate cmd wizard on a real console The cmd e2e CI step still asserted the OLD contract (installer writes ~/.claude/commands/plannotator-*.md with the ! prefix). Commands are dead; the step now guards the NEW contract: a fresh install must write NO plannotator command files. The Gemini TOML assertions stay — they now verify the verbatim checkout copy delivers intact files (the tag the e2e pins, v0.17.1, contains apps/gemini/commands). The failure also exposed that install.cmd ran the wizard on redirected stdin (set /p falls through to defaults at EOF) — and Q3's default being yes meant CI silently ran `npm install -g glimpseui`. cmd now probes for a real console via `timeout /t 0` (errors when stdin is redirected), matching sh's /dev/tty and ps1's IsInputRedirected gates: no console, no wizard, no wizard-only installs.	2026-06-05 09:23:45 -07:00
Michael Ramos	ac2630078f	feat: add PLANNOTATOR_GLIMPSE env var to disable Glimpse native window ... Add glimpse config option to PlannotatorConfig and resolveUseGlimpse() resolver. Wire into both Bun and Pi browser-opening paths. Document PLANNOTATOR_GLIMPSE, PLANNOTATOR_GLIMPSE_WIDTH, and PLANNOTATOR_GLIMPSE_HEIGHT in env var reference and AGENTS.md. For provenance purposes, this commit was AI assisted.	2026-06-03 14:19:31 -07:00
Ashwin John Chempolil	6a7fd415a7	Add Kiro CLI integration (skills + custom agent, cross-platform installers) (#837) ... Universal, auto-detected Kiro CLI support — no flag, no separate installer. When ~/.kiro exists (or kiro-cli is on PATH), the installer copies Plannotator skills and a custom agent into ~/.kiro, the same convention used for Codex and Gemini. - packages/shared: add the kiro-cli agent origin (badge, prompt runtime, PLAN_TOOL_NAMES). Origin is cosmetic for Kiro (no dedicated AI provider). - apps/kiro-cli: 3 origin-baked skills (review/annotate/archive) + an example custom agent that wires every skill via skill:// resources and a plannotator-scoped shell tool. setup-goal + visual-explainer install from apps/skills (no content duplication). - scripts/install.sh: auto-detect ~/.kiro, sparse-checkout apps/kiro-cli, install 3 kiro + 2 shared skills + the agent (never clobbering an existing one); covered by scripts/install.test.ts. - scripts/install.ps1 + install.cmd: Windows parity mirroring the Codex pattern. Statically verified only; not yet runtime-tested on Windows. - docs: installation + Kiro guides, AGENTS.md, env-var reference updated. Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-06-02 11:04:18 -07:00
Michael Ramos	8c947c5419	Add Amp plugin integration (#803) ... * Add Amp plugin integration * Use official Amp logo on landing page * Tighten landing agent selector * Default landing selector to Claude Code * Stabilize server ready handoff test * Create ready handoff directory before writing * Stabilize server ready handoff tests * Fix Amp command cancellation and cwd * Preserve Plannotator browser handling for Amp * Fix Amp review edge cases * Add PowerShell installer smoke coverage	2026-05-27 21:49:30 -07:00
Hrand Liu	e0aee7451b	feat: add PLANNOTATOR_DATA_DIR env var to customize data directory (#795) ... * feat: add PLANNOTATOR_DATA_DIR env var to customize data directory * fix: update missed hardcoded paths to use PLANNOTATOR_DATA_DIR OpenCode plugin and VS Code extension still used hardcoded ~/.plannotator paths, causing the IPC registry and plan backing file to diverge from the server when PLANNOTATOR_DATA_DIR is set. Also exports data-dir from @plannotator/shared and documents the new env var in AGENTS.md. Co-authored-by: Chris Werner Rau <14326070+cwrau@users.noreply.github.com> Co-authored-by: João O. Santos <34689526+Joao-O-Santos@users.noreply.github.com> * fix: vendor data-dir.ts into Pi extension and rewrite imports The Pi extension copies shared/server modules into generated/ at build time. Without vendoring data-dir.ts and rewriting the parent-relative imports, typecheck fails on all generated files that import getPlannotatorDataDir. * refactor: eliminate duplicated data-dir logic and clean up call sites - VS Code extension: replace inlined getPlannotatorDataDir() copy with import from the canonical packages/shared/data-dir.ts (esbuild bundles it, so no runtime dependency needed) - storage.ts: hoist repeated getPlannotatorDataDir() calls to a module-level DATA_DIR constant, matching the pattern config.ts uses - data-dir.ts: remove inaccurate docstring claim about relative path resolution (the code does not call resolve()) - improvement-hooks.ts: hoist to DATA_DIR constant, clarify comments on the two-level hook lookup (hooks/ subdir vs root fallback) * fix: resolve relative PLANNOTATOR_DATA_DIR to absolute path A relative value like ./data would break readArchivedPlan's path traversal guard, which compares a resolve()'d absolute path against the still-relative planDir prefix. Always return an absolute path so all callers get consistent path shapes. * fix: use @plannotator/shared/data-dir imports in server package Switch from relative ../shared/data-dir imports to the package export, matching the convention every other server file follows. Update Pi vendor script sed rules to match the new import style. * fix: use package imports in server and respect data dir in compound skill Server modules: switch from relative ../shared/data-dir imports to @plannotator/shared/data-dir, matching the convention every other server file follows. Update Pi vendor script sed rules to match. Compound skill: update hardcoded ~/.plannotator paths to check PLANNOTATOR_DATA_DIR first, so the skill reads plans and writes the improvement hook to the correct location when users set a custom data directory. Co-authored-by: Chris Werner Rau <14326070+cwrau@users.noreply.github.com> Co-authored-by: João O. Santos <34689526+Joao-O-Santos@users.noreply.github.com> * fix: remove remaining hardcoded ~/.plannotator assumptions - Settings UI: replace hardcoded path in label and placeholder with generic text that doesn't assume a specific data directory - quickLabels: update agent tip to reference PLANNOTATOR_DATA_DIR so the agent checks the correct plans directory - codex-review: hoist getPlannotatorDataDir() to module-level DATA_DIR constant, eliminating redundant per-call resolution in debugLog() - Tests: make submit-plan and storage tests resilient to PLANNOTATOR_DATA_DIR being set in the environment - Install scripts (sh, ps1, cmd): check PLANNOTATOR_DATA_DIR before falling back to ~/.plannotator for config.json attestation lookup * fix: expand tilde in install script and update test assertions install.sh: PLANNOTATOR_DATA_DIR set to ~/... stays literal inside double quotes, so the config file check silently failed. Add case statement to expand ~ the same way the runtime data-dir.ts does. install.test.ts: update three assertions that checked for hardcoded ~/.plannotator paths — now verify PLANNOTATOR_DATA_DIR awareness instead. * docs: add PLANNOTATOR_DATA_DIR to env var reference with VS Code note Document the new env var on the marketing site's environment variables reference page. Include a footnote about ensuring VS Code inherits the variable when launched from the Dock. --------- Co-authored-by: Michael Ramos <mdramos8@gmail.com> Co-authored-by: Chris Werner Rau <14326070+cwrau@users.noreply.github.com> Co-authored-by: João O. Santos <34689526+Joao-O-Santos@users.noreply.github.com>	2026-05-26 14:56:07 -07:00
Michael Ramos	2a5528693f	Add Ask AI to plan and annotate reviews (#763) ... * Add Ask AI to plan and annotate reviews - mount shared AI runtime on plan, annotate, review, and Pi servers - add shared document chat UI and comment-popover Ask AI entry point - default providers from detected agent origin and document first-run announcement - reuse shared code-review AI hook and document setup/docs * Address Ask AI review followups Stabilize AI chat session resets and callbacks, preserve per-origin provider defaults, share AI chat formatting utilities, and avoid blocking startup on model discovery. * Fix AI permission details and Pi CLI lookup Show formatted tool input in document AI permission cards and make Pi AI provider CLI detection work on Windows. * Fix Pi model discovery in AI capabilities Keep AI runtime startup non-blocking while making the capabilities endpoint await pending Pi/OpenCode model discovery before returning provider metadata.	2026-05-21 01:32:17 -07:00
Michael Ramos	95750ab657	feat: search-based code navigation with peek view (#711) ... * feat: search-based code navigation with peek view (#694) Add IDE-like code navigation to the review UI. Cmd/Ctrl+click a token in a diff to find its definitions and references across the repo via ripgrep, displayed in a VS Code-style peek panel below the diff. Backend: bounded rg search with language-aware definition patterns (TS/JS, Python, Go, Rust), ranked results (same file > changed files > same directory), confidence labels, and graceful degradation when rg is not installed. Both Bun and Pi servers implement the endpoints. Frontend: Dockview peek panel with syntax-highlighted full-file preview on the left and grouped reference list on the right. Clicking a reference scrolls the preview; double-clicking an in-diff result navigates to the file with a gold line flash. Closes #694 * fix: resolve Pi server TypeScript errors for code-nav Add missing spawn import, type the close callback parameter, and use double-cast for parseBody → CodeNavRequest. * fix: clear loading state on cached preview hits Without this, clicking a cached file while a fetch is in-flight leaves isLoading stuck true — the spinner hides the preview. * feat: show pointer cursor on Cmd/Ctrl+hover for navigable tokens Adds pn-token-nav class with thicker underline and pointer cursor when hovering a token while holding the modifier key, signaling the token is Cmd+clickable for code navigation. * refactor: remove go-to-diff navigation from peek panel Strip the in-diff badge, double-click-to-jump, highlightDiffLine wiring, and onCodeNavGoToDiff from the peek panel. The peek view is the primary interaction — jump-to-diff adds complexity without clear value at this stage. * chore: remove dead code from code-nav cleanup Delete unused highlightDiffLine.ts, remove codeNavChangedFiles and codeNavActiveSide from context and App.tsx, drop stale extractChangedFiles import. * chore: add code-nav endpoints to AGENTS.md, remove dead activeSide state * fix: don't classify bare indented calls as definitions Change the TS/JS method pattern from zero-or-more (*) to one-or-more (+) declaration keywords, so plain calls like startServer(config) are no longer misclassified as definitions. * feat: show toast when code-nav is unavailable in platform-only PR mode Instead of opening the peek panel and showing misleading "No results", Cmd+click in non-local PR mode shows a brief toast explaining that code navigation requires a local checkout. * fix: strip hljs hardcoded background from code-nav preview The highlight.js github-dark theme sets a fixed dark background on all .hljs elements. Apply transparent override for the entire peek preview so code inherits the active theme's background.	2026-05-13 12:24:37 -07:00
Michael Ramos	1eb561551c	feat: standalone skills package + HTML render-annotate mode (#687) ... Add --render-html flag to plannotator annotate that renders HTML files as-is in an iframe instead of converting to markdown. Includes annotation support via postMessage bridge, sharing via paste service, and theme inheritance from Plannotator's 30+ themes. New skill: plannotator-visual-explainer — wraps nicobailon/visual-explainer with Plannotator theme tokens, extended patterns (timelines, SVG diagrams, code blocks, risk tables, Pierre diffs via CDN), and plan/PR-specific guidance. All three servers (Bun, Pi, OpenCode) support the new flag.	2026-05-11 10:30:49 -04:00
Michael Ramos	f4493fc6ff	Optimize plan editor rendering (#696) ... - Replace blocks/frontmatter useState + useEffect with useMemo (eliminates mount cascade) - Migrate toast notifications from React state to Sonner (App no longer re-renders for toasts) - Extract AppHeader component with React.memo (header no longer re-renders on unrelated state changes) - Fix color transition flash on initial load (suppress global * transition until mount settles) - Add Tailwind @source path for new editor components directory	2026-05-11 08:55:18 -04:00
Michael Ramos	13c667c044	feat(hook): PFM reminder & improvement hook support across all runtimes (#689) ... PFM reminder & improvement hook support across Claude Code, OpenCode, and Pi. - Add opt-in PFM reminder (pfmReminder config flag) injected on EnterPlanMode - Wire composeImproveContext() into all three runtimes - Fix OpenCode system.transform array reference bug (pushes were going to dead array) - Fix install scripts silently stripping PreToolUse/EnterPlanMode hook entry - Isolated Pi sandbox testing (--no-extensions -e)	2026-05-11 08:14:13 -04:00
Graeme Folk	69ef11bdfb	feat(review): add jj review workflows (#675) ... * feat(review): add jj support for local diffs * feat(review): add jj review workflows * fix(review): tighten jj diff defaults * test(review): add jj manual sandbox * fix(review): share jj agent diff prompts * fix(review): quote jj agent revsets * feat(review): share jj vcs handling with pi * fix(review): tighten jj bookmark and pi pr handling * fix(review): tighten jj defaults and detection * fix(review): harden jj diff and vcs detection --------- Co-authored-by: Michael Ramos <mdramos8@gmail.com>	2026-05-07 19:57:33 -07:00
Michael Ramos	84a0b434f9	fix(ui): smart resolution + existence-validation for code-file paths (#654) ... * fix(ui): smart resolution + existence-validation for code-file paths The bare-prose / backtick path detector linkifies anything that looks like a code path. Two failure modes regularly produce dead links: prose abbreviations like `editor/App.tsx` (real file is `packages/editor/App.tsx`) and references to files the plan proposes but hasn't created yet. Both 404 on click with no UX cue. Resolves abbreviated paths via a case-insensitive suffix-match against a cached project walk (`resolveCodeFile` in `packages/shared/resolve-file.ts`), mirroring what `resolveMarkdownFile` already does for markdown. The walk is pre-warmed when the plan/annotate server boots and on every `/api/doc` request, with a 30s TTL so newly-created files can resolve mid-review. Storing the walk as a Promise makes the cache race-safe — concurrent callers piggyback rather than starting a second walk. A new `POST /api/doc/exists` endpoint takes a batch of candidate paths and reports `found` / `ambiguous` / `missing` / `unavailable` per path. On the frontend, `useValidatedCodePaths` extracts candidates from the markdown on load and POSTs once. The renderer reads the result via `CodePathValidationContext`: `found` opens directly with the resolved absolute path, `ambiguous` opens a `CodeFilePicker` popover listing all matches (common in monorepos where `App.tsx` exists in several packages), `missing` demotes the link to plain code, and `unavailable` falls back to the optimistic linkification we have today. While validation is in flight, every detected path renders as a link, so first paint is unchanged. The detection itself gets a shape filter (`isPlausibleCodeFilePath`) that hard-rejects shell brace expansion (`{a,b}`), glob wildcards, and whitespace, while explicitly allowing `[` / `]` so Next.js dynamic routes (`app/[slug]/page.tsx`) still resolve. The bare-prose regex moves out of `InlineMarkdown.tsx` into `code-file.ts` so the renderer and the new server-side extractor use the same source of truth, and the extractor strips fenced code blocks, HTML comments, and URL ranges before scanning so it only emits candidates the renderer would actually paint. Pi extension mirrors the Bun changes (handler upgrade, pre-warm, `/api/doc/exists` route). When the popout's `/api/doc` request 404s the dialog now surfaces "File not found in repo: <path>" instead of silently swallowing the error. Tests: `code-file.test.ts` extended with shape-filter and Next.js-route cases; new `extract-code-paths.test.ts` covers extraction, dedup, fenced/HTML/URL exclusion, and the URL-with-parens regression; new `resolve-file.test.ts` covers the suffix-match strategy, leading `./` handling, ambiguous results, and ignored-dir behavior. * fix(ui): thread doc-base through code-path validator Out-of-tree linked docs (and annotate-mode files outside cwd) reference files relative to themselves. The validator was resolving against cwd only, so those paths got marked missing and the renderer demoted them to plain text — even though clicks still resolved correctly via base. Also tightens the suffix-match's leading-segment strip so `../foo.ts` no longer silently misresolves to an unrelated `foo.ts` in cwd. Cleanup: delete unused extract-code-paths import in reference-handlers, add the export entry to packages/shared so consumers don't rely on Bun's lenient subpath resolution. Add TODO(security) comments at both handleDocExists sites flagging that absolute paths bypass project-root containment. 223 tests pass (3 new resolver cases for baseDir + ../ regression). * refactor(editor): dedupe activeDocBaseDir; expand security TODO Self-review fallout: 1. The doc-base expression `linkedDocHook.filepath ? dirname(...) : imageBaseDir` lived in two places (click-time URL builder and Viewer prop). If they drift, validator and click resolve against different bases and we silently re-introduce the demote-correct-link bug. Extract to a single useMemo. 2. The handleDocExists security TODO mentioned absolute paths in `paths[]` but I just added `base` acceptance, which has the same shape of leak (hostile sender supplies base=/secret/dir + relative path). Both vectors flagged in one TODO, mirrored Bun + Pi. 223 tests pass; both builds clean. * fix(ui): code-file popout shows real error; misc consistency Review fallout: - `CodeFilePopout` hardcoded "File not found in repo" regardless of cause. The hook already captures the server's error string, so an ambiguous-path 400 (which can happen if a user clicks an optimistic link before validation completes) was surfacing as a misleading not-found message. Render the actual `error` and only show the planned/future-file caveat when the error matches "file not found". - `InlineMarkdown` emitted demoted bare-prose paths as raw strings while every other plain-text branch in `emitPlainTextWithBareUrls` routes through `transformPlainText`. Cosmetic-only today since paths rarely contain transformable content, but the divergence invites copy-paste rot. Routed through the same helper. - CLAUDE.md missed the new POST /api/doc/exists endpoint in both Plan Server and Annotate Server tables. Added. 223 tests pass; both builds clean. * fix(ui): demote paths the extractor excluded from validation When the validator is ready but a candidate path has no entry in the validated map, the extractor intentionally excluded it — e.g. inside an HTML comment or fenced code block. The renderer was optimistically linking these because gateCodePath returned 'link' for missing entries. Found during manual testing: `<!-- packages/editor/App.tsx -->` inside a paragraph (parser doesn't recognize HTML comments as block-level) was rendered as a clickable link. The extractor correctly stripped the comment, but the renderer's optimistic fallback overrode that. Also adds manual test harness: tests/manual/path-detection/ with sandbox setup + three launcher scripts (plan mode, annotate in-tree, annotate out-of-tree) covering ~30 test cases. 223 tests pass; both builds clean. * fix(ui): skip HTML comments in InlineMarkdown scanner The parser doesn't recognize <!-- --> as block-level HTML, so comments inside paragraphs fall through to InlineMarkdown. The scanner then finds paths inside the comment text and linkifies them. The previous gateCodePath fix (demote when not in validated map) didn't help here because the same path appeared elsewhere in the document — the map had an entry from the non-comment occurrence. Fix: match <!-- ... --> at the top of the scanner loop and skip the entire comment. HTML comments should be invisible per CommonMark spec. * fix: handle unavailable variant in markdown resolve narrowing The shared ResolveResult type gained an `unavailable` variant for code files. The markdown resolver never returns it, but TS can't narrow past it without an explicit guard. Both Bun and Pi handlers now guard `not_found || unavailable` before accessing `result.path`.	2026-05-04 14:21:16 -07:00
Michael Ramos	5bc57fc5bd	feat(ui): shortcut registry foundation (#652) ... * feat(ui): shortcut registry foundation Salvages the foundation of #380 by @gjermundgaraba — the original PR sat 6 weeks waiting for review and was closed by the author. This lands the engine, scope data, and auto-generated marketing docs page, deferring per-component handler migration to follow-up PRs. Layout: - packages/ui/shortcuts/{core,runtime,index}.ts — engine + barrel - packages/ui/shortcuts/plan-review/* — scopes for plan editor surfaces - packages/ui/shortcuts/code-review/* — scopes for review editor surfaces - packages/{editor,review-editor}/shortcuts.ts — per-app surface composition - apps/marketing/src/{components,lib} — auto-generated docs page App behavior is unchanged: existing ad-hoc keydown handlers keep running. The registry is plumbing waiting for migration PRs to wire to it. Co-authored-by: Gjermund Garaba <gjermund@garaba.net> * feat(ui): add missing shortcuts; warn on hold-binding gap Addresses code review on #652. Adds five live shortcuts that were missing from the registry, so the auto-generated /docs/reference/keyboard-shortcuts page is no longer incomplete: - Mod+P (print plan) — plan review + annotate - Mod+B (toggle file tree) — code review (new "Layout" section) - Mod+. (toggle review sidebar) — code review - Mod+Shift+T (toggle demo tour, dev-only) — code review - Mod+Shift+F (focus PR comments search) — new code-review/prComments scope Also whitelists `.` as a valid normalized binding token in the registry validator (keeps catching typos like Cmd/Ctrl while allowing legitimate punctuation keys). Hold-binding gap: the registry's `Alt hold` advertises a shortcut that `useShortcutScope` cannot dispatch. Today this is harmless because the existing `useInputMethodSwitch` hook handles it bespoke. Added warning comments at the binding declaration and next to `useDoubleTapShortcuts` so future migrators don't fall in. A `useHoldShortcuts` hook is the proper fix and will land with the App.tsx migration that needs it. * feat(ui): comprehensive shortcut registry coverage Two-Sonnet inventory audit found 19 more user-facing shortcuts not in the registry, on top of the 5 added in the previous commit. Adding all of them so the auto-generated /docs/reference/keyboard-shortcuts page ships honest and the in-app help modal (once migrated to read from the registry) is comprehensive. Plan / annotate surface (4 added): - Escape closes image lightbox (Viewer) - Mod+Enter saves annotation edit (AnnotationPanel) - Escape cancels annotation edit (AnnotationPanel) - Enter confirms image name (ImageAnnotator name field) Code review surface (15 added): - v / a in single-file diff mode (toggle viewed / stage) - v, a, x, z, c, [, ] in all-files (new AllFilesDiff scope) - Mod+Enter / Escape / Tab (new SuggestionModal scope) - Mod+Enter / Escape (new AI Assistant scope: AITab + AskAIInput) - Escape closes tour dialog (new TourDialog scope) Engine: extended NAMED_TOKENS with `[` and `]` for the all-files navigation bindings. Tests updated for the new section ordering. App behavior unchanged — this is pure declarative metadata. The existing ad-hoc keydown handlers keep running; the registry is plumbing for the migration follow-up PRs. * fix(ui): suggestion modal scope advertised non-existent submit shortcut Self-review caught: SuggestionModal.tsx only has Escape (close) and Tab (indent) handlers — no Mod+Enter. The Mod+Enter the inventory agent referenced lives in AnnotationToolbar's suggested-code textarea, which is already covered by reviewAnnotationToolbarShortcuts.submitComment. Removed the bogus submit action from the scope. Renamed cancel→close since the modal preserves suggestion state on Escape (it's not a discard). * fix(review): remove broken Cmd+Shift+C copy-diff shortcut The handler at App.tsx:695 checked `e.key === 'c'` but Shift forces `e.key` to uppercase `'C'`, so the shortcut never fired. The in-app help modal also advertised it incorrectly as "Toggle comment mode". Removing the broken shortcut entirely rather than fixing the case mismatch — the toolbar's Copy button already does the same thing, and there's no evidence anyone was relying on the keyboard shortcut. Three deletions: - packages/review-editor/App.tsx — handler block - packages/review-editor/shortcuts.ts — copyDiff registry action - packages/ui/components/KeyboardShortcuts.tsx — help modal entry * chore(ui): address review nits - F3: refresh AGENTS.md scope inventory (added scopes were missing) - F5: getShortcutPlatform() now delegates to packages/ui/utils/platform isMac instead of duplicating the regex+SSR guard - F7: drop the stand-alone imageAnnotator.confirmName entry; the behaviour is folded into the existing save action's hint so the marketing docs page no longer renders two Enter rows with different descriptions - F8: remove the unused listenerOptions parameter from useShortcutScope (no callers; inline objects in the dep array would have caused re-render loops if anyone tried to use it) TODOs left for the migration follow-up: - F2: cross-scope arbitration — dispatchShortcutEvent should bail on event.defaultPrevented; needs preventDefault to be claimed consistently across scopes first - F6: matchesKeyToken doesn't actually match the Space token; needs a special case before any scope binds Space App behaviour unchanged. --------- Co-authored-by: Gjermund Garaba <gjermund@garaba.net>	2026-05-04 10:15:31 -07:00
Michael Ramos	9b7c39d2a4	fix(review): move hide-whitespace to server-side git diff -w (#635 follow-up) (#638) ... The client-side approach from PR #635 normalized file contents before diffing, which destroyed all indentation. Move whitespace handling to the server by threading a `-w` flag through the git diff pipeline. - Add `GitDiffOptions` to review-core.ts, inject `-w` in all 7 diff type paths + untracked file diffs - Thread options through git.ts → vcs.ts → review.ts / Pi server - Read `hideWhitespace` from ~/.plannotator/config.json on startup so the initial diff already respects the persisted preference - Accept `hideWhitespace` in `/api/diff/switch`, echo in responses - Client toggles trigger a lightweight server refetch that preserves the active file (no panel reset) - Handle edge case where current file disappears when `-w` removes whitespace-only diffs - Remove broken client-side parseDiffFromFile/normalize hack - Update API docs in AGENTS.md Closes the indentation bug reported by @zeroZshadow on PR #631. For provenance purposes, this commit was AI assisted.	2026-04-30 17:23:55 -07:00
Michael Ramos	bb404f8d14	feat: stacked PR review — PR switching, scope toggling, multi-PR posting (#620) ... * feat(shared): add isSameProject, PR stack types, and PR list provider Extends PRRef/PRMetadata with defaultBranch, PRStackInfo, PRStackTree, PRStackNode, PRDiffScope, and PRListItem types. Adds isSameProject() for owner/repo validation on PR switching. Adds fetchPRStack() and fetchPRList() dispatch functions (GitHub-only for now, GitLab stubs). Includes 9 new tests for isSameProject covering GitHub, GitLab, and cross-platform scenarios. For provenance purposes, this commit was AI assisted. * feat(shared): add GitHub PR stack tree walking and PR list fetching Implements fetchGhPRStack() which walks up/down the PR stack via GraphQL, resolving numbers and titles for each node in the chain. Collapses queryPRsByHead/queryPRsByBase into a single queryPRsByRef helper. Adds fetchGhPRList() using gh pr list. Fixes GHE support by removing hostnameArgs from fetchGhPRList (--repo already handles GHE). Filters jq "null" string from defaultBranch detection. For provenance purposes, this commit was AI assisted. * feat(shared): fetch defaultBranch for GitLab MRs Queries the project's default_branch via glab API so getPRStackInfo can detect stacked MRs on GitLab. Best-effort — caught errors fall back to undefined. For provenance purposes, this commit was AI assisted. * feat(shared): add PR stack detection and full-stack diff module New pr-stack module with: - getPRStackInfo(): detects stacked PRs from baseBranch vs defaultBranch - getPRDiffScopeOptions(): generates layer/full-stack scope options - runPRFullStackDiff(): computes diff from default branch to HEAD - resolvePRFullStackBaseRef(): resolves origin/main or local main - checkoutPRHead(): fetches and checks out a PR head in a worktree - buildMinimalStackTree(): builds UI tree from stack info Includes 13 tests covering ref resolution, branch fallbacks, and GitLab ref formats. For provenance purposes, this commit was AI assisted. * feat(shared): add worktree pool for per-PR agent isolation Creates a session-scoped pool of git worktrees — each PR visited during a stacked review gets its own isolated checkout. Agents run in their PR's worktree undisturbed by PR switches. Handles deduplication of concurrent ensure() calls for the same PR. Includes 11 tests covering caching, cross-repo restrictions, GitLab ref formats, and cleanup. For provenance purposes, this commit was AI assisted. * feat(shared): add diffScope/prUrl to agent jobs, branch diff type Adds prUrl and diffScope optional fields to AgentJobInfo so agent findings carry the PR and scope context they were launched under. Exports new pr-stack and worktree-pool modules from package.json. Adds 'branch' to DefaultDiffType union for branch diff as default. For provenance purposes, this commit was AI assisted. * feat(ui): add PR annotation fields, Popover, and SearchableSelect Extends CodeAnnotation with prUrl, prNumber, prTitle, prRepo, and diffScope fields for stacked PR attribution. Adds shared Popover wrapper around radix-ui. Adds SearchableSelect for filterable dropdown lists (used by PR selector). For provenance purposes, this commit was AI assisted. * feat(ui): add branch diff as default option, new Git settings tab Adds 'Branch' as a fourth default diff type option in both the first-run dialog and settings panel. Moves the default diff type setting from the Display tab to a new Git tab in review mode. Updates config store validators to accept 'branch'. For provenance purposes, this commit was AI assisted. * feat(server): add prUrl/diffScope plumbing to agent jobs and prompts Threads prUrl and diffScope through the agent job lifecycle so findings carry the PR and scope they were generated under. Adds full-stack prompt branch to codex-review and tour-review — when in full-stack mode, the diff is inlined in the prompt instead of telling the agent to run git diff. Re-exports isSameProject and new PR provider functions from server/pr.ts. For provenance purposes, this commit was AI assisted. * feat(server): add stacked PR support to Bun review server Adds PR switching, layer/full-stack scope toggling, PR list caching, worktree pool integration, and multi-PR platform posting to the Bun review server. Key additions: - /api/pr-diff-scope: switch between layer and full-stack diffs - /api/pr-list: cached PR list for the current repo - /api/pr-switch: in-place navigation between PRs in a stack - /api/pr-action: targetPrUrl support for multi-PR posting - /api/file-content: full-stack branch for hunk expansion - prSwitchCache/prStackTreeCache for session-scoped caching - diffScope tagging on agent job completion - Scope guard: returns 400 on full-stack diff failure instead of overwriting the working diff with empty content For provenance purposes, this commit was AI assisted. * feat(ai): pass cwd to Claude agent SDK for worktree support Forwards the working directory to the Claude agent provider so agents run in the correct worktree when reviewing stacked PRs. For provenance purposes, this commit was AI assisted. * feat(pi): add stacked PR support to Pi server (Bun parity) Mirrors all stacked PR features from the Bun server: - PR switching, scope toggling, PR list, multi-PR posting - prSwitchCache/prStackTreeCache with initial PR seeding - diffScope/prUrl plumbing in agent jobs - Worktree pool creation and lifecycle - Full-stack file-content resolution matching Bun's guard structure - targetPrUrl support on /api/pr-action Hoists worktreePool declaration to outer scope in plannotator-browser to fix TS18004 scoping error. Updates vendor.sh for new shared modules. For provenance purposes, this commit was AI assisted. * feat(hook): create worktree pool for PR review sessions Creates a worktree pool when opening a PR review with --local, seeding it with the initial PR's checkout. Integrates pool cleanup into server shutdown. Passes the pool to startReviewServer for agent isolation during PR switching. For provenance purposes, this commit was AI assisted. * feat(review-editor): add hooks for PR stack, context, and annotations - useAnnotationFactory: stamps prUrl/prNumber/prTitle/prRepo/diffScope onto annotations, only when viewing a stacked PR - usePRStack: handles scope selection and PR switching with loading state - usePRContext: adds URL-change detection to prevent stale-fetch race when switching PRs (discards in-flight responses for previous PR) For provenance purposes, this commit was AI assisted. * feat(review-editor): add stacked PR UI components - PRSelector: searchable dropdown for switching between PRs in a repo - PRSwitchOverlay: loading animation during PR switch - StackedPRLabel: stack tree popover with scope selector and PR navigation - ReviewSubmissionDialog: multi-PR submission dialog with per-target status, orphaned findings section with copy-as-markdown, and partial failure retry For provenance purposes, this commit was AI assisted. * feat(review-editor): multi-PR export with heading hierarchy Updates exportReviewFeedback for multi-PR sessions: - Groups annotations by prUrl, then by file within each PR - Uses proper heading hierarchy (## for files, ### for annotations in multi-PR mode) - Detects single-PR mismatch (annotations from a different PR than the current view) and uses annotation-level PR context - Adds diffScope labels per PR group when present Includes 5 new tests: multi-PR headings, single-PR mismatch, diffScope labels, and non-stacked annotation handling. For provenance purposes, this commit was AI assisted. * feat(review-editor): integrate stacked PR into sidebar, diff panel, and agents - ReviewSidebar: groups annotations by PR in multi-PR sessions, shows PR headers with annotation counts - ReviewDiffPanel: filters annotations by prUrl and diffScope so only matching annotations appear in the diff gutter - ReviewStateContext: adds prDiffScope to shared review state - ReviewAgentJobDetailPanel: shows diffScope in job detail - PRSummaryTab: shows stack info in PR summary - index.css: PR switch shimmer and overlay animations For provenance purposes, this commit was AI assisted. * feat(review-editor): wire stacked PR into main review app Integrates all stacked PR features into the review editor: - PR stack state management (prStackInfo, prStackTree, prDiffScope) - applyPRResponse: shared handler for PR switch and scope toggle, preserves active file index on scope changes - Multi-PR platform posting via Promise.allSettled with parallel requests, partial failure retry, and per-target status tracking - ReviewSubmissionDialog replaces inline dialog JSX - useAnnotationFactory stamps PR context onto annotations - keepalive on /api/feedback to survive tab closure - Proper try/catch/finally on handlePlatformAction For provenance purposes, this commit was AI assisted. * docs: add stacked PR review documentation Updates AGENTS.md, code-review command docs, and AI code review guide with stacked PR review capabilities. For provenance purposes, this commit was AI assisted. * fix(server): stamp prNumber/prTitle/prRepo on agent findings Agent annotations only had prUrl and diffScope, missing prNumber, prTitle, and prRepo. When agent findings were the only annotations for a PR target in the submission dialog, the target rendered as #0 with no title. Now resolves full PR context from prSwitchCache at job completion and stamps all five fields. Both Bun and Pi. For provenance purposes, this commit was AI assisted. * feat(ui): rename diff options — "Committed" replaces "Branch" / "Current PR Diff" Consolidates two confusing committed-diff options into one: - Settings/first-run: "Committed" — "Everything you've committed on this branch" - Mid-session switcher: "Committed changes" (replaces both "vs main" and "Current PR Diff") Uses merge-base under the hood (matches GitHub PR behavior). Removes the two-dot branch diff from the UI — it stays in the runtime DiffType union for backwards compat. Old "branch" values in config/cookies are silently upgraded to merge-base. Git settings tab now uses radio cards with descriptions instead of a cramped segmented control. First-run dialog descriptions rewritten in plain language — no git commands. For provenance purposes, this commit was AI assisted. * fix(review-editor): rename client-side "PR Diff" labels to "Committed changes" DiffTypePicker.tsx had a hardcoded "PR Diff" override for merge-base when the base picker is present. exportFeedback.ts also used "PR Diff" in export labels. Both now say "Committed changes" to match the server-side label and settings UI. For provenance purposes, this commit was AI assisted. * fix(server): discover stack UI for root PRs targeting the default branch Root PRs (baseBranch === defaultBranch) were excluded from stack detection because getPRStackInfo returned null. Now the server always fetches the stack tree in PR mode. If the tree reveals descendant PRs, prStackInfo is retroactively set with source "tree-discovered", enabling the stack UI, scope selector, and PR navigation from the root of a stack. Both Bun and Pi servers updated. Adds "tree-discovered" to the PRStackInfo source union. For provenance purposes, this commit was AI assisted. * fix(review-editor): derive diff scope from annotations, not UI state The export function now reads diffScope from annotations instead of the prReviewScope parameter. Fixes two issues: 1. Agent job "Copy All" showed the wrong scope when the user switched between layer/full-stack after launching the agent 2. Mixed-scope annotations produced a confusing "layer, full-stack" comma-joined label instead of grouping by scope Extracts renderScopedGroups helper for scope-aware grouping — used by both single-PR and multi-PR export paths. When annotations share one scope, it appears in the header. When mixed, annotations are grouped under ## Layer / ## Full-stack headings. Includes 4 new tests: uniform scope derivation, mixed scope grouping, single scope header, and prReviewScope override prevention. For provenance purposes, this commit was AI assisted. * fix(server): add tree-discovered stack fallback to pr-switch handler The initial-load path upgrades prStackInfo for root PRs when the stack tree reveals descendants, but the pr-switch handler was missing this logic. The server now sends correct prStackInfo after switching to a root-of-stack PR. Both Bun and Pi. Also removes stale prReviewScope dependency from agent job panel's copyAllText useMemo. For provenance purposes, this commit was AI assisted. * fix: extract resolveStackInfo helper, fix stack UI on non-stacked PRs Extracts the tree-discovered stack fallback into resolveStackInfo() in pr-stack.ts — eliminates 4 copies of the same logic across Bun startup, Bun pr-switch, Pi startup, and Pi pr-switch. Fixes StackedPRLabel showing on every PR: the check now counts non-default-branch nodes (> 1) instead of all nodes (> 1). Without this, every PR showed a "Stack (1 PR)" popover because the tree always has at least [defaultBranch, currentPR]. For provenance purposes, this commit was AI assisted. * fix(review-editor): don't re-open already-succeeded PR tabs on retry On partial failure retry, openUrls was pre-seeded with URLs from previously succeeded targets, causing those PR pages to re-open in the browser alongside newly succeeded ones. Now starts empty — only URLs from the current attempt are opened. For provenance purposes, this commit was AI assisted. * refactor(review-editor): extract PR session state into usePRSession hook Consolidates 5 independent useState calls (prMetadata, prStackInfo, prStackTree, prDiffScope, prDiffScopeOptions) into a single usePRSession hook with atomic updatePRSession callback. Replaces two identical 5-line setter blocks (initial load and applyPRResponse) with single updatePRSession calls. All ~60 consumer sites unchanged — same variable names via destructuring. For provenance purposes, this commit was AI assisted.	2026-04-27 22:07:55 -07:00
Michael Ramos	d102c5f709	feat(annotate): add --gate, --json, and --silent-approve flags (#570) ... Adds an opt-in review gate flow to annotation mode with three composable flags: - `--gate`: 3-way UX (Approve / Send Annotations / Close) - `--json`: structured decision output (`{"decision":"approved|annotated|dismissed"}`) - `--silent-approve`: suppresses plaintext approve marker for naive hooks Includes shared arg parser, @-reference handling, updated templates across all harnesses (Claude Code, Copilot, Gemini, OpenCode, Pi), and full documentation. Closes #570 For provenance purposes, this commit was AI assisted.	2026-04-23 17:40:40 -07:00
Michael Ramos	68972aa6a6	feat(review): custom base branch for code review (#599) (#602) ... Adds a searchable base-branch picker to the code review UI so reviewers can compare against any branch, not just the auto-detected default. Key changes: - Base-branch picker (Radix Popover + search, local/remote groups) - Diff-type picker (Radix DropdownMenu + info tooltips) - Worktree picker (Radix Popover + search) - getDefaultBranch prefers upstream ref (origin/main) for fresher diffs - listBranches populates GitContext.availableBranches - resolveBaseBranch trusts caller verbatim (no silent rewrites) - --end-of-options hardens git argv against flag-injection via base refs - Branch diff / PR Diff always available (removed hide-on-default guard) - merge-base agent prompt runs git itself instead of inlining the patch - Agent jobs snapshot diff context at launch for honest "Copy All" - Feedback export includes diff mode + base header for agent context - Round-trip base through /api/diff so refresh preserves the selection - selectedBase / committedBase split prevents file-content race - Optimistic picker reverts on switch failure - DiffViewer remounts on mode switch to invalidate stale file content - Worktree switch preserves active diff mode - Pi parity: initialBase, currentBase, agent diffContext, vendor.sh - Tooltip wide mode, TooltipProvider mounted, keyboard/touch fixes - GPT-5.5 added to Codex model picker - CLAUDE.md API table updated Closes #599 For provenance purposes, this PR was AI assisted.	2026-04-23 16:49:25 -07:00
Michael Ramos	289a2aa494	chore: bump version to 0.19.0 ... Bump version across all 7 package/plugin manifests. Also refresh AGENTS.md (new Block types from #597, Code Tour endpoints from #569), prune removed flags from the Pi README (--plan-file, /plannotator-set-file from #595), and pin Renovate off bun-version bumps to keep the macOS codesign hotfix from v0.17.9 in place. For provenance purposes, this commit was AI assisted.	2026-04-21 19:36:02 -07:00
Michael Ramos	3a7f1156cd	chore: bump version to 0.18.0 ... Also fix stale annotate description in README and add missing PLANNOTATOR_ORIGIN entry to the env vars table. For provenance purposes, this commit was AI assisted.	2026-04-19 17:41:53 -07:00
Michael Ramos	b780739291	feat(annotate): support HTML files and URL annotation (#545) ... * fix(annotate): sanitize dangerous link protocols in markdown renderer Block javascript:, data:, and vbscript: URLs in InlineMarkdown link rendering. Links with dangerous protocols render as plain text instead of clickable anchors. Uses a blocklist approach so existing links with custom protocols (obsidian://, vscode://, Windows C:\ paths) continue to work. For provenance purposes, this commit was AI assisted. * feat(annotate): add HTML-to-markdown and URL-to-markdown utilities - html-to-markdown.ts: Turndown wrapper with GFM table rule, strips script/style/noscript tags - url-to-markdown.ts: Jina Reader (free, returns markdown) with fetch+Turndown fallback. Warns on Jina failure, auto-skips Jina for local/private URLs (localhost, 192.168.*, 10.*, etc.) - config.ts: add jina setting and resolveUseJina() with priority chain --no-jina flag > PLANNOTATOR_JINA env > config.json > default true For provenance purposes, this commit was AI assisted. * feat(annotate): support HTML files and URLs in annotate command Extend the annotate subcommand to accept .html/.htm local files (converted via Turndown) and https:// URLs (fetched via Jina Reader with fetch+Turndown fallback). URL content is fetched terminal-side before opening the browser. Add --no-jina global flag to disable Jina Reader per-invocation. Add 10MB file size guard for local HTML files. For provenance purposes, this commit was AI assisted. * feat(annotate): HTML files in folder browser and on-demand conversion - Widen file browser glob to include .html/.htm alongside markdown - handleDoc converts HTML files via Turndown on demand when selected - hasMarkdownFiles accepts optional extensions param for folder validation - Add sourceInfo field to annotate server API response - Add _site/, public/, out/, .docusaurus/, .jekyll-cache/, storybook-static/ to FILE_BROWSER_EXCLUDED For provenance purposes, this commit was AI assisted. * feat(annotate): source attribution badge for HTML/URL annotations Show a subtle badge in DocBadges displaying the URL hostname or HTML filename for converted content. Thread sourceInfo from API response through App → Viewer → DocBadges. Also update Pi extension to accept HTML-only folders in annotate mode. For provenance purposes, this commit was AI assisted. * test: update CLI help text assertion for HTML/URL annotate support For provenance purposes, this commit was AI assisted. * fix(annotate): address PR review findings Security: - Add project-root containment check for HTML files in /api/doc handler using exported isWithinProjectRoot() from resolve-file.ts - Blocks path traversal via absolute paths or ../ escapes isLocalUrl fixes: - Add bracketed IPv6 loopback [::1] detection - Replace hostname.startsWith('10.') with proper IPv4 regex to avoid matching public hostnames like 10.example.com Revert Pi extension change: - Pi server doesn't implement HTML file browsing or conversion yet - Keep Pi folder validation markdown-only until both implementations are updated per CLAUDE.md guidelines Cleanup: - Remove dead el.children || el.childNodes fallback in table rule - Extract hostnameOrFallback() helper to @plannotator/shared/project replacing duplicated try/catch IIFEs in DocBadges and index.ts For provenance purposes, this commit was AI assisted. * feat(annotate): Pi extension HTML annotation parity Bring the Pi extension to full parity with the Bun server for HTML annotation support: - Vendor html-to-markdown and url-to-markdown via vendor.sh - walkMarkdownFiles now scans .html/.htm alongside markdown - handleDocRequest converts HTML files on-demand via Turndown with isWithinProjectRoot containment check - serverAnnotate includes sourceInfo in /api/plan response - index.ts supports URL detection (Jina Reader + fallback), HTML file detection with Turndown conversion, folder HTML validation, and 10MB file size guard - openMarkdownAnnotation accepts and threads sourceInfo - Add turndown as a Pi extension dependency For provenance purposes, this commit was AI assisted. * fix(pi): Obsidian vault walks stay markdown-only, add try/catch for HTML - Add extensions param to walkMarkdownFiles (default: HTML-inclusive) - Obsidian callers pass /\.mdx?$/i to match Bun server behavior - Add try/catch around HTML file reads in handleDocRequest For provenance purposes, this commit was AI assisted. * fix(annotate): address second review — base-block traversal, metadata IP, dead code Security: - Add isWithinProjectRoot check to the base-relative block for HTML files in both Bun and Pi /api/doc handlers. Previously HTML files served via the base query param bypassed the containment guard. - Add 169.254.0.0/16 (link-local / cloud metadata) to isLocalUrl private IP ranges Cleanup: - Remove dead hostname === "[::1]" check (WHATWG URL parser strips brackets; hostname === "::1" already handles it) - Remove dead parent?.childNodes fallback in table cell() function For provenance purposes, this commit was AI assisted. * refactor(annotate): replace custom table rules with turndown-plugin-gfm Drop ~60 lines of hand-rolled GFM table conversion that had a bug (tables without explicit <thead> produced invalid GFM). Use the official turndown-plugin-gfm plugin (24KB) which correctly handles all table patterns plus adds strikethrough and task list support. For provenance purposes, this commit was AI assisted. * fix(annotate): handle all CommonMark backslash escapes in InlineMarkdown Expand the backslash escape regex to cover all CommonMark-defined escapable characters (. ) - # > + | { } &), not just the subset the parser uses for formatting. Fixes literal backslashes appearing in rendered output for Turndown-escaped content like "1\." → "1.". For provenance purposes, this commit was AI assisted. * fix(annotate): prevent SSRF via redirect to private/local URLs Replace redirect: "follow" with redirect: "manual" in fetchViaTurndown and validate each redirect hop against isLocalUrl. Blocks attacks where an external URL redirects to cloud metadata endpoints (169.254.169.254) or other private IPs. Limits redirect chain to 10 hops. For provenance purposes, this commit was AI assisted. * chore: update lockfile for turndown-plugin-gfm in Pi extension bun install needed to resolve turndown-plugin-gfm in the Pi extension workspace after adding it to apps/pi-extension/package.json. For provenance purposes, this commit was AI assisted. * fix(annotate): switch to @joplin/turndown-plugin-gfm, fix TS errors Replace unmaintained turndown-plugin-gfm (2017, v1.0.2) with the actively maintained Joplin fork (2025, v1.0.64, 16KB). Fix TypeScript errors that broke CI: - Add @ts-expect-error for untyped @joplin/turndown-plugin-gfm import - Restructure fetchViaTurndown redirect loop to avoid uninitialized variable — first fetch before loop, loop only for redirects For provenance purposes, this commit was AI assisted. * fix(annotate): use proper declarations.d.ts instead of ts-expect-error Add declarations.d.ts for @joplin/turndown-plugin-gfm with typed function signatures, remove the ts-expect-error suppression. For provenance purposes, this commit was AI assisted. * fix: explicitly include declarations.d.ts in shared tsconfig CI's tsc wasn't finding the ambient module declaration with implicit include. Add explicit include to ensure declarations.d.ts is always picked up regardless of environment. For provenance purposes, this commit was AI assisted. * fix: use ts-expect-error for @joplin/turndown-plugin-gfm types CI's tsc does not pick up ambient declarations.d.ts files despite local tsc finding them — likely a module resolution discrepancy between environments. Revert to @ts-expect-error which passes in both CI and local typecheck. For provenance purposes, this commit was AI assisted. * fix(annotate): body size limit for URL fetches, redirect error, file: protocol - Add 10MB body size limit to both Jina and fetch+Turndown URL paths, matching the local HTML file guard. Streams response body and aborts if limit exceeded. - Distinguish "Too many redirects" from a genuine 3xx response after redirect loop exhaustion. - Add file: to the dangerous protocol blocklist in sanitizeLinkUrl. For provenance purposes, this commit was AI assisted. * fix(annotate): HTML folder outside cwd, HTML linked doc navigation - Remove containment check from base-relative block for HTML files in both Bun and Pi /api/doc handlers. Matches markdown behavior so HTML files in annotated folders outside cwd are served correctly. Standalone block (no base) retains its cwd check as fallback. - Widen isLocalMd → isLocalDoc to treat .html/.htm links as linked documents. Clicking [Next](next.html) in a converted page now opens it via /api/doc with Turndown conversion instead of a new browser tab. For provenance purposes, this commit was AI assisted. * fix(annotate): full loopback range, drain redirect bodies, document env vars - Expand loopback check from just 127.0.0.1 to the full 127.0.0.0/8 range so all loopback addresses skip Jina Reader - Cancel redirect response body before re-fetching to avoid leaking TCP connections back to the pool - Document PLANNOTATOR_JINA and JINA_API_KEY in CLAUDE.md env var table For provenance purposes, this commit was AI assisted. * fix(annotate): IPv6 loopback, readBodyWithLimit fallback, env var docs, comments - Add [::1] back to isLocalUrl — WHATWG URL hostname getter preserves brackets for IPv6 (verified: Bun and Node both return "[::1]"). Add comment explaining the empirical verification so future reviewers don't re-flag. - Fix readBodyWithLimit null-body fallback to still enforce the 10MB limit via text length check instead of silently falling through. - Document PLANNOTATOR_JINA and JINA_API_KEY in AGENTS.md env var table (CLAUDE.md is a symlink to AGENTS.md). - Add comments to base-relative blocks in both Bun and Pi handleDoc explaining the intentional lack of containment check (matches pre-existing markdown behavior, base is set server-side). For provenance purposes, this commit was AI assisted. * fix(annotate): block IPv4-mapped IPv6 and private IPv6 ranges in isLocalUrl Add PRIVATE_IPV6 regex matching bracketed IPv6 private/reserved ranges: - ::ffff: (IPv4-mapped — embeds private IPv4 as hex, e.g. [::ffff:c0a8:1]) - fe80: (link-local) - fc00::/7 (unique-local, covers fc00:: through fdff::) Closes the redirect-SSRF bypass where a public URL redirects to a private address expressed as IPv4-mapped IPv6, e.g. http://[::ffff:169.254.169.254]/latest/meta-data/ For provenance purposes, this commit was AI assisted. * fix(annotate): document IPv6 hostname verification, sourceInfo type, annotate flow - Expand isLocalUrl comment with full empirical verification table showing actual hostname getter output for every IPv6 format in both Bun and Node — prevents false-positive review findings about brackets - Add sourceInfo to /api/plan response type in App.tsx for type safety - Update CLAUDE.md annotate flow diagram to reflect HTML/URL/folder input types For provenance purposes, this commit was AI assisted. * fix(annotate): escape \(, cancel response bodies on error, doc sourceInfo - Add ( to backslash escape regex alongside existing ) — Turndown emits \( in link-adjacent contexts - Cancel response body before throwing on !res.ok in both fetchViaJina and fetchViaTurndown error paths (redirect loop already did this) - Document sourceInfo field in AGENTS.md annotate server API table For provenance purposes, this commit was AI assisted. * fix(annotate): skip base injection for URL annotations, body cleanup - Skip dirname(filePath) base injection when filePath is a URL in both Bun and Pi annotate servers. dirname on a URL string produces a nonsensical filesystem path, causing linked doc clicks to 404. URL annotations now let links open normally instead. - Cancel response body before throwing on content-type mismatch and content-length overflow in fetchViaTurndown/readBodyWithLimit. - Fix double parseInt in readBodyWithLimit content-length check. - Correct AGENTS.md flow diagram: OpenCode not yet implemented for HTML/URL annotation. For provenance purposes, this commit was AI assisted. * feat(annotate): OpenCode HTML file and URL annotation support Add URL detection (Jina Reader + fallback), HTML file detection with Turndown conversion, 10MB file size guard, and sourceInfo threading to OpenCode's handleAnnotateCommand. Uses the same shared utilities as the Bun CLI and Pi extension. OpenCode uses the Bun server directly (startAnnotateServer from @plannotator/server/annotate), so no server-side changes needed — only the command handler routing was missing. Note: folder annotation mode is not added (OpenCode didn't have it before this PR for markdown either — separate scope). For provenance purposes, this commit was AI assisted. * chore(annotate): update slash command description, align fetch log messages - OpenCode plannotator-annotate.md description now mentions HTML/URL - Align fetch progress messages across all three clients: all now show "(via Jina Reader)" or "(via fetch+Turndown)" consistently For provenance purposes, this commit was AI assisted. * fix(annotate): skip conversion for .md URLs, wikilink HTML targets, cleanup - URLs ending in .md/.mdx are fetched raw — no Jina, no Turndown. Content is already markdown. Removes text/plain from fetchViaTurndown content-type whitelist since .md URLs are now short-circuited. - Wikilink regex widened to preserve .html/.htm targets instead of appending .md (e.g. [[page.html]] no longer becomes page.html.md) - Remove redundant existsSync before statSync in OpenCode handler For provenance purposes, this commit was AI assisted. * test(annotate): add htmlToMarkdown conversion tests Tests cover the core conversion utility that all three clients depend on: - Basic HTML → markdown (headings, paragraphs, links, code blocks) - Tables with and without <thead> (the GFM plugin bug that was caught) - Script/style/noscript stripping - Strikethrough (GFM) - Empty HTML handling - Dangerous links preserved (sanitization is in the renderer, not here) For provenance purposes, this commit was AI assisted. * fix(annotate): check content-type before treating .md URLs as raw markdown URLs ending in .md/.mdx (e.g. GitHub's viewer page for README.md) may return HTML instead of raw markdown. fetchRawText now checks the response content-type — if the server returns HTML, returns null so the caller falls through to Jina/Turndown for proper conversion. For provenance purposes, this commit was AI assisted. * fix(annotate): add SSRF redirect protection to fetchRawText fetchRawText (for .md/.mdx URLs) was using default redirect: "follow" with no isLocalUrl validation on redirect hops — a .md URL redirecting to 169.254.169.254 would be followed and credentials returned as "markdown". Now uses redirect: "manual" with per-hop isLocalUrl checks, matching fetchViaTurndown's SSRF protection. For provenance purposes, this commit was AI assisted.	2026-04-12 18:56:28 -07:00
Michael Ramos	ed254cf3bd	Supply-chain hardening: version pinning, SLSA attestations, fix #506 (#512) ... * feat(install): supply-chain hardening (#507) + fix Gemini crash on Windows (#506) Closes #507 items 1-2 and fixes #506. Issue #507 asked for version-pinned installs from a trusted source, immutable releases, and build-provenance attestations. This commit ships: - `--version v0.X.Y` / positional / `-Version` flag across all three installers (bash, PowerShell, cmd) so users can pin to reviewed versions. Default stays `latest` — every existing `curl | bash` / `irm | iex` invocation works unchanged. install.cmd gained `--version` as an alias to its existing positional form for surface-area consistency. - SLSA build provenance attestations via `actions/attest-build-provenance` (SHA-pinned to v4.1.0) in release.yml. Covers all 10 compiled binaries (5 plannotator + 5 paste-service). Top-level workflow permissions tightened from `contents: write` to `contents: read`, with per-job overrides where needed. Attestation step is gated on tag pushes so PR dry-runs don't pollute Sigstore's transparency log. - Opt-in provenance verification in all three installers, resolved via a three-layer precedence ladder: 1. CLI flag `--verify-attestation` / `-VerifyAttestation` 2. Env var `PLANNOTATOR_VERIFY_ATTESTATION=1` 3. Config `~/.plannotator/config.json` → `verifyAttestation: true` 4. Default off Off-by-default matches every major ecosystem installer (rustup, brew, bun, deno, helm) and avoids a UX failure for the majority of users who don't have `gh` installed or authenticated. Security-conscious users get three ergonomic opt-in paths. When enabled, the installer hard-fails if `gh` is missing so opt-in is never silently skipped. - `PlannotatorConfig.verifyAttestation?: boolean` added to `packages/shared/config.ts`. Pure additive schema change; no runtime consumer exists (the field is read only by the install scripts). No UI surface — this is an OS-level power-user knob only. - Documentation rewritten in README.md, apps/hook/README.md, and the marketing install page with both pinned-version examples and a "Verifying your install" section covering manual `gh attestation verify` and offline `cosign verify-blob` paths. - Release skill updated to note that the release pipeline now emits SLSA attestations and (post-merge) GitHub Immutable Releases will be enabled on the repo. Issue #506 (install.cmd crash on Windows when Gemini is present): Root cause was cmd.exe's `setlocal enabledelayedexpansion` eating `!` chars in the embedded `node -e "..."` Gemini settings merge script. Cmd's Phase 2 parser treated `!s.hooks)s.hooks={};if(!` as a variable expansion, corrupting the JS before node saw it. Fixed by rewriting the JS to use the `||` idiom (`s.hooks = s.hooks || {}`) which contains no `!` characters — semantically identical, and sidesteps cmd's parser entirely with no escape gymnastics and no new dependencies. Added regression test in scripts/install.test.ts that asserts the `||` form is present and `if(!s.hooks)` is absent, so re-introducing the bug would fail CI. Test suite expanded from 23 to 29 tests covering: - Gemini #506 regression - Three-layer opt-in wiring assertions for all three installers - install.sh guard check (the executable `gh attestation verify` call must live behind the `verify_attestation -eq 1` gate, so the default path never invokes gh) - PlannotatorConfig schema assertion Out of scope (tracked): - Immutable Releases toggle in repo Settings (one-click, post-merge). - SLSA Build L3 via `slsa-framework/slsa-github-generator` (requires restructuring release.yml around a reusable workflow). - Issue #507 item 3: UI update-cooldown setting. For provenance purposes, this commit was AI assisted. * fix(install): address PR #512 review feedback - docs: drop broken `cosign verify-blob` example. Verified against Sigstore's docs that `cosign verify-blob` requires `--bundle` (or `--signature` + `--certificate`); the shipped command provided neither and could not have worked. Replaced with the canonical `gh attestation verify --repo` flow (gh + auth required, but optional — only needed if a user wants to manually audit provenance) and a link to GitHub's offline-verification docs for advanced workflows. Applied to README.md, apps/hook/README.md, and the marketing install page. - docs: list per-platform binary paths in the manual verification examples. The previous snippets hardcoded `~/.local/bin/plannotator` even though install.ps1 writes to `%LOCALAPPDATA%\plannotator\` and install.cmd writes to `%USERPROFILE%\.local\bin\` — Windows users copying the snippet got file-not-found instead of a verification result. - install.sh: capture and surface `gh attestation verify` stderr on failure instead of redirecting to /dev/null. Diagnosability matches install.ps1, which already prints `$verifyOutput` on failure. The most common failure mode (`gh auth login` not run) is now immediately actionable instead of presenting as a generic "verification failed" error. - install.cmd: reject any unknown dash-prefixed token before the positional fall-through. A typoed `--verify-attesttion` no longer becomes `VERSION=--verify-attesttion` and 404s on a nonsensical download URL — it fails fast with `Unknown option:` and a usage hint. install.sh and install.ps1 already had equivalent guards (case `-*)` arm and PowerShell's strict param block respectively). All 29 install tests still pass. For provenance purposes, this commit was AI assisted. * test(ci): add Windows integration job for install.cmd Address the reviewer gap flagged in PR #512: the unit tests in scripts/install.test.ts only do file-content string matching on Linux and never execute cmd.exe, so the #506 fix (rewriting the embedded `node -e` Gemini merge to use `x = x || {}` instead of `if(!x)x=...`) was shipped without any runtime coverage. The previous CI only had ubuntu-latest runners. New `install-cmd-windows` job on `windows-latest`: 1. Seeds a fake `~/.gemini/settings.json` with a pre-existing non- plannotator hook plus unrelated top-level keys (theme, general). The fixture mirrors the shape of a real Gemini settings.json but uses only obviously-fake values and contains no secrets. 2. Runs `scripts\install.cmd v0.17.1 --skip-attestation` end-to-end through real cmd.exe. This exercises the parser under `enabledelayedexpansion`, the embedded `node -e` merge script, and the full install flow. 3. Parses the post-install settings.json with PowerShell and asserts: - The plannotator hook was added to hooks.BeforeTool. - The pre-existing fixture hook is still present (regression guard for the original #506 bug, where cmd ate the `!` in `if(!s.hooks.BeforeTool)s.hooks.BeforeTool=[]` and wiped existing arrays). - Unrelated top-level keys (`theme`, `general.ciFixtureSentinel`) survived the merge untouched. 4. Separately exercises the new unknown-flag rejection added in the previous commit: invokes `install.cmd --verify-attesttion` (typo) via Start-Process and asserts exit code != 0. Before the review fix this would have silently set `VERSION=--verify-attesttion` and 404'd on the download. The job runs in parallel with the existing ubuntu `test` job (no deps, independent runner). Uses the v0.17.1 release as the binary source — that release is pre-PR, so the test is stable against release drift and is testing install.cmd's CODE, not any specific binary. This closes the CI gap where install.cmd had effectively zero runtime coverage and the original #506 bug could have recurred without anyone noticing until a user reported it. For provenance purposes, this commit was AI assisted. * fix(install.cmd): capture gh stderr on failure (consistency with install.sh) Self-review catch: the PR #512 reviewer flagged install.sh for redirecting `gh attestation verify` output to /dev/null, which swallowed actionable error messages (auth missing, network issue, attestation not yet propagated) behind a generic "verification failed" line. I fixed install.sh in the previous review-fix commit but missed that install.cmd had the exact same pattern: gh attestation verify "!TEMP_FILE!" --repo !REPO! >/dev/null 2>&1 Same bug, same consequence, same fix. cmd doesn't have bash's `$(cmd)` or PowerShell's `& cmd 2>&1` output capture, so we redirect to a temp file and `type` it on failure, then clean up in both branches: gh attestation verify ... > "%TEMP%\gh-output.txt" 2>&1 if !ERRORLEVEL! neq 0 ( type "%TEMP%\gh-output.txt" >&2 del "%TEMP%\gh-output.txt" echo Attestation verification failed! >&2 ... ) del "%TEMP%\gh-output.txt" All three installers now surface gh's actual error message on failure, which makes the most common failure mode (`gh auth login` not run) immediately diagnosable on every platform. Note: this code path is not exercised by the new Windows CI integration job because that job passes `--skip-attestation`, and exercising the gh verify path would require an attestation for v0.17.1 to exist — which it doesn't, since v0.17.1 was released before this PR added the attestation step. The fix will first become CI-testable against the first post-merge release that carries a provenance bundle. For provenance purposes, this commit was AI assisted. * fix(install): address second review pass on PR #512 Five findings, all verified against the actual code: - **AGENTS.md env var table** (reviewer: "CLAUDE.md"; it's a symlink to AGENTS.md) was missing PLANNOTATOR_VERIFY_ATTESTATION. Added with an explicit note that it's read by the install scripts only, not the runtime binary — since every other entry in that table is a runtime env var, the distinction matters. - **install.cmd unknown-flag guard metacharacter injection.** The previous guard ran `echo %~1 | findstr /b "[-]"`, where %~1 is unquoted before the pipe. A user passing `install.cmd "--bad&calc"` would have cmd expand %~1 to `--bad&calc`, see the `&` as a command separator, and execute `calc` as a side effect before the flag check. Not a remote exploit (user already has shell exec), but a defensive coding weakness in supply-chain hardening code. Replaced with a variable-assigned substring test using delayed expansion — `set "CURRENT_ARG=%~1"` preserves metacharacters literally inside the `"..."` set syntax, and `!CURRENT_ARG:~0,1!` extracts the first char without any subprocess. This also fixes the same bug in the error-message echo, which previously echoed the unquoted `%~1` and re-triggered metacharacter interpretation in the error path itself. The echo now uses `"%~1"`. Note: the reviewer's proposed one-liner `if "%~1:~0,1%"=="-"` was syntactically invalid — cmd's `:~start,length` substring modifier does not work on positional parameters, only on regular variables. A variable assignment is necessary. - **install.cmd unquoted --repo argument** in the `gh attestation verify` call. TEMP_FILE was quoted but REPO was not. REPO is hardcoded to `backnotprop/plannotator` so not exploitable, but inconsistent with install.sh (which quotes `"$REPO"`). One-char fix: `--repo "!REPO!"`. - **test.yml intentional-typo drift hazard.** The unknown-flag regression test invokes `install.cmd --verify-attesttion` (missing an `a`). The only assertion was `$p.ExitCode -eq 0`. If a future typo-sweep "fixes" the misspelling to the valid `--verify-attestation`, install.cmd would accept the flag, proceed to download the latest release, run `gh attestation verify` against it, and — because v0.17.1 pre-dates the attestation step — fail with a different non-zero exit. Both paths exit 1, so the test would silently drift from "guard works" to "gh attestation verify fails on pre-PR release" without anyone noticing. Two-part fix: 1. Explicit comment marking the misspelling as intentional ("do not correct during a typo sweep"). 2. Redirect stderr to a temp file and assert it contains "Unknown option:" — the actual discriminator between the guard triggering and any other failure mode that happens to exit non-zero. - **README.md verification block was too long** for the main README. Trimmed from ~33 lines (intro + 3 code blocks + opt-in mechanisms + precedence notes) to a single sentence that links to the canonical marketing installation docs where the full content already lived. Same treatment applied to apps/hook/README.md for consistency. The marketing docs are unchanged and remain the single source of truth for verification workflows. All 29 install tests still pass. The Windows CI integration job's new stderr assertion will exercise the harder guard on the next push. For provenance purposes, this commit was AI assisted. * fix(install): tighten attestation verify with --source-ref and --signer-workflow Addresses PR #512 review cycle 3 finding that repo-scoped verification alone doesn't bind the downloaded binary to the specific tag the user requested. A misattached release asset would pass the old check because the wrong binary would still carry a valid attestation for its own (wrong) commit. GitHub's own docs explicitly recommend both constraints: "The more precisely you specify the identity, the more control you will have over the security guarantees. Ideally, the path of the signer workflow is also validated." — https://cli.github.com/manual/gh_attestation_verify All three installers now pass: --source-ref "refs/tags/<requested-tag>" Enforces that the git ref the attestation was produced from matches the tag the installer asked for. Closes the misattached-asset gap. --signer-workflow backnotprop/plannotator/.github/workflows/release.yml Enforces that the attestation was signed by our release workflow file specifically, not any workflow in the repo. GitHub treats this flag as a regex (see cli/cli#9507) so future refactors can broaden the match without breaking version-pinned installs to historical releases. Also addresses the sibling finding that install.cmd used a fixed %TEMP%\gh-output.txt temp filename while the rest of the script uses %RANDOM% for uniqueness. Renamed to %TEMP%\plannotator-gh-%RANDOM%.txt, matching the established pattern and removing a theoretical race between concurrent invocations. New test in install.test.ts asserts all three installers pass --source-ref and --signer-workflow with the expected values. 30 tests pass. For provenance purposes, this commit was AI assisted. * fix(install): address PR #512 review cycle 4 (parser edges, ps1 stream, docs) Five findings, all verified against actual code. One bonus fix in install.sh for a sibling bug the reviewer flagged only in install.cmd. install.ps1: `Write-Host $verifyOutput` on attestation failure wrote gh's diagnostic to PowerShell's Information stream (stream 6), which is silently dropped when CI pipelines capture stderr. Replaced with `[Console]::Error.WriteLine($verifyOutput)` — direct stderr handle, matches the behavior of `echo ... >&2` in install.sh and `type ... >&2` in install.cmd. install.sh + install.cmd: `--version --some-other-flag` used to set VERSION to the flag name (e.g. VERSION=--verify-attestation), which then tried to download tag `v--verify-attestation` and 404'd. The empty-check on `$2`/`%~2` didn't catch dash-prefixed values. Added an explicit dash-prefix check that returns a clean "--version requires a tag value, got flag: X" error instead of degrading into a cryptic download failure. install.sh + install.cmd: mixing `--version v1.0.0 stray` used to silently overwrite VERSION with "stray" because the positional branch unconditionally assigned VERSION=$1. Added a VERSION_EXPLICIT sentinel that's set to 1 when --version is seen, and the positional branch now errors with "Unexpected positional argument: X (version already set)" when it sees a token while the sentinel is set. Same sentinel is also set by the positional branch itself, so passing two positional version tokens also errors out cleanly. Note: the reviewer flagged the positional-overwrite bug only in install.cmd, but install.sh had the identical issue (same unconditional `VERSION="$1"` in the `*)` arm) and the same dash- check gap in both its `--version <val>` and `--version=<val>` branches. Fixing both installers symmetrically — inconsistency here would just trigger another review round. marketing/installation.md: the "Verifying your install" prose promised a "cryptographic link to the exact commit and workflow run," but the example commands only passed `--repo`, which just proves the artifact came from some workflow in our repository. The installer now constrains with `--source-ref` and `--signer-workflow` after review cycle 3, so the docs were out of sync with the actual installer behavior. Updated all three platform examples (bash, pwsh, cmd) to include the tighter flags with a placeholder (`vX.Y.Z`) and a sentence explaining what the extra flags actually buy the user. README.md and apps/hook/README.md are already link-only after cycle 2 and don't need changes. install.test.ts: two new tests. - Regression guard asserts install.sh and install.cmd contain the VERSION_EXPLICIT sentinel, the dash-prefix error message, and the "Unexpected positional argument" guard. Anyone removing any of these in a future cleanup would fail CI. - Regression guard asserts install.ps1 uses [Console]::Error.WriteLine and does NOT use Write-Host for verifyOutput. 32 tests pass (was 30). Smoke-tested install.sh with `--version --verify-attestation` and `--version v1.0.0 stray` — both now exit 1 with clean usage errors instead of silent download failures. For provenance purposes, this commit was AI assisted. * fix(install): address PR #512 review cycle 5 Five code/doc fixes, all verified against actual code. Finding 1 from the review (opt-in verification unusable until a post-merge release is cut) is correct but not actionable — it's inherent to how SLSA attestations work and the only "fix" is timing + release cadence. install.ps1: `[Console]::Error.WriteLine($verifyOutput)` silently converted multi-line gh output to the literal string "System.Object[]" — the opposite of what cycle 4's Write-Host fix was supposed to do. `& gh ... 2>&1` captures multi-line output as an object[] array; passing the array directly to [Console]::Error.WriteLine binds to the WriteLine(object) overload and calls ToString() on the array. Fixed by piping through Out-String first (and TrimEnd to drop the trailing newline it adds). Confirmed against Sigstore/PowerShell docs and the Delft Stack array-to-string guide. install.cmd: replaced `echo !TAG! | findstr /b "v"` with a substring test `if not "!TAG:~0,1!"=="v"`. Same metacharacter-injection class as the parser bug fixed in cycle 2 — piping an unquoted expanded variable re-exposes cmd's & | > < operators in the value before the pipe runs. Inconsistent to leave this one instance using the unsafe pattern when every other comparable check in the script uses the substring idiom. install.cmd: randomized the two remaining deterministic temp file paths — %TEMP%\release.json and %TEMP%\plannotator-<tag>.exe — to match the %RANDOM% pattern already used by GH_OUTPUT. Closes two gaps at once: concurrent-invocation collisions (real for automated upgrade tooling) and same-user symlink pre-placement (the SHA256 check passes on authentic content, but a symlink at the predictable path would redirect where curl writes the binary before the install move runs). All three installers: reject --verify-attestation and --skip-attestation together as mutually exclusive instead of trying to guess which the user meant. Previously install.sh/cmd took last- on-command-line wins and install.ps1 took a fixed-priority Skip- always-wins (documented but inconsistent with the other two). No sane user passes both flags — fast-failing with a clear "mutually exclusive" error is better than silently picking one and hoping it matches intent. Guards live inline in both arms of the bash/cmd parsers and right after the PowerShell param block. test.yml: added a comment block on the install.cmd v0.17.1 pin explaining why that version was chosen, why `latest` isn't used, what the prerequisites are for bumping it, and what failure mode to expect if the pinned release is ever removed. No behavior change — the existing pin stays. Addresses the reviewer's concern that the dependency was undocumented. install.test.ts: four new regression guards. - Asserts install.ps1 uses Out-String (not bare [Console] call on raw $verifyOutput) for multi-line gh output - Asserts all three installers reject the --verify+--skip combo with a "mutually exclusive" error and install.ps1 has the `$VerifyAttestation -and $SkipAttestation` guard - Asserts install.cmd uses randomized temp paths for release.json and the binary download, and that the old deterministic paths are gone - Asserts install.cmd uses the substring test for v-prefix normalization and does not pipe echo|findstr for that check 35 install tests pass (was 32). Smoke-tested the bash mutex guard in both orders — both fail fast with "mutually exclusive" and exit 1 regardless of which flag appears first. For provenance purposes, this commit was AI assisted. * fix(install.cmd): randomize checksum temp path + tighten test assertions Self-review catch on top of the cycle 5 commit: - `%TEMP%\checksum.txt` (lines 164/172/174) was still a fixed predictable path. Same concurrency + symlink-pre-placement class as release.json and TEMP_FILE that cycle 5 fixed. Inconsistent to fix two of three and leave the third. Renamed to `%TEMP%\plannotator-checksum-%RANDOM%.txt` matching the established pattern. The reviewer didn't flag this one — I missed it during the cycle 5 sweep. - Tightened the Out-String regression test from a weak "Out-String appears somewhere in the file" check to a regex matching the specific `$verifyOutput | Out-String` wiring. Previous assertion would have passed even if some future bug accidentally wrapped the mutex-guard string literal in Out-String while leaving $verifyOutput unprotected. - Expanded the randomized-temp-paths test to cover all four curl download targets (release.json, binary, checksum sidecar, gh output capture) rather than the two originally in scope, and to assert the old fixed paths (including checksum.txt) are gone. 35 tests still pass. For provenance purposes, this commit was AI assisted. * fix(install.cmd): escape ! in Claude Code slash command files Pre-existing bug flagged in PR #512 review cycle 6. install.cmd writes the three Claude Code slash command files (plannotator-review.md, plannotator-annotate.md, plannotator-last.md) via `echo` lines inside `setlocal enabledelayedexpansion`. cmd.exe's Phase 2 parser strips unmatched `!` characters — so lines like: echo !`plannotator review $ARGUMENTS` ended up in the written file as: `plannotator review $ARGUMENTS` without the leading `!`. The `!` prefix is what tells Claude Code to execute the backtick block as a shell command; without it, Claude Code renders the line as inline markdown code and the slash command is a silent no-op. The install appeared to succeed, but every Windows cmd user got three broken slash command files. install.sh (single-quoted heredocs) and install.ps1 (single-quoted here-strings) write the `!` correctly because their respective literal-string idioms bypass shell expansion entirely. install.cmd has no single-quote-literal equivalent — its escape hatch is `^!`. The Gemini section of install.cmd (lines 482, 495) already uses `^!` correctly; the Claude Code section didn't until now. Fix: three characters — `echo !` → `echo ^!` on lines 334, 351, 368. Brings install.cmd into parity with the other two installers. No divergence introduced; existing divergence removed. Two regression guards added: - Unit test in install.test.ts asserts install.cmd contains the escaped form for all three command files and does not contain the unescaped form. - New step in the Windows CI integration job reads back each generated .md file from %USERPROFILE%\.claude\commands\ and asserts it contains the literal `!`\`plannotator` prefix. Catches the bug at the actual file-write level on a real Windows runner, not just via source-code grep. 36 install tests pass (was 35). Note: the broader architectural issue — all three installers carry hand-typed duplicates of command content that already lives at apps/hook/commands/*.md — is deferred to a follow-up issue. The cmd bug is the visibly-broken symptom; the deduplication is the long-term fix. For provenance purposes, this commit was AI assisted. * fix(install.cmd): double-caret escape for ! in slash command echoes The previous fix used `echo ^!` for the three Claude Code slash command files. The Windows CI integration job's new file-readback assertion proved this is wrong: the generated plannotator-review.md still landed with no `!` prefix, making the slash command a silent no-op as before. Root cause: cmd has two escape phases under enabledelayedexpansion. Phase 1 (parse time): `^` escapes the next char. `^!` → `!`. The caret is consumed. Phase 2 (delayed expansion): the remaining bare `!` is an unmatched variable reference and gets stripped. Single `^!` dies in Phase 2 because Phase 1 already ate the caret. Double `^^!` survives: Phase 1 reduces `^^` to `^` (leaving `^!`), Phase 2 treats the caret as an escape for `!` and emits a literal. Cycle 6's fix got the direction right but the arithmetic wrong. The new file-readback assertion in test.yml caught it on the first real CI run, which is exactly why that assertion was added. Also fixes the Gemini slash command echoes (lines 482, 495) which used the identical incorrect `^!` pattern. The review comment flagged Gemini as "correct" based on source-reading alone; there was never any CI coverage for the Gemini file contents, and the Gemini section was silently broken for the same reason. Both sections now use `^^!`. Unit test updated to assert the double-caret form on all five echo lines (three Claude Code, two Gemini) and reject both the unescaped and single-caret variants. For provenance purposes, this commit was AI assisted. * fix(install.ps1): fall back to x64 on ARM64 Windows instead of 404ing Pre-existing bug surfaced in PR #512 review cycle 7. install.ps1 detected ARM64 correctly and set $arch=arm64, constructing a URL for plannotator-win32-arm64.exe — which doesn't exist in any release. The release pipeline only builds bun-windows-x64 (release.yml line 88), so there is no native ARM64 Windows binary to download. With $ErrorActionPreference=Stop set at the top of the script, the resulting 404 on Invoke-WebRequest threw a terminating error and the install aborted with a stack trace. ARM64 Windows PowerShell users could not install plannotator at all. Meanwhile install.cmd, which hardcodes PLATFORM=win32-x64 and lets ARM64 hosts pass the arch check, silently installs the x64 binary and relies on Windows 11's x86-64 emulation layer to run it. This is accidentally the useful behavior — imperfect, but the user gets a working install instead of a hard failure. This commit brings install.ps1 into parity with install.cmd's (accidentally correct) behavior: - On 64-bit Windows, $arch is unconditionally "x64" — no more branch for arm64 that would download a nonexistent binary. - When PROCESSOR_ARCHITECTURE == ARM64, Write-Host prints a notice telling the user they're getting the x64 binary via Windows emulation so the behavior isn't silent. - 32-bit Windows still errors out (unchanged). Both Windows installer paths now produce a working install on both x64 and ARM64 hosts. No release pipeline changes. No new binaries. The test `detects ARM64 architecture` used to be a weak string- presence check that passed whether the ARM64 branch selected arm64 or x64. Rewrote it to assert the actual new contract: ARM64 is detected (for the notice), $arch is hardcoded to "x64", and the previous `{ "arm64" }` branch is gone so the regression can't silently return. Native ARM64 Windows builds tracked as a follow-up — requires verifying Bun's Windows ARM64 target support and adding bun-windows-arm64 to the release matrix. For provenance purposes, this commit was AI assisted. * fix(install): pre-flight MIN_ATTESTED_VERSION guard + placeholder docs PR #512 cycle 7 review surfaced that opt-in provenance verification was dead-on-arrival for the window between this PR merging and the first post-merge release: - The docs showed `--version v0.17.1` as the pinned example. v0.17.1 was cut before this PR added attestation generation to release.yml, so any user copy-pasting the example AND enabling verification would hit a cryptic `gh: no attestations found` error and a hard install failure. - Default installs with verification enabled (via flag, env var, or config file) resolve `latest` to v0.17.1 and hit the same failure with no user-visible pinned version to "blame." Medium fix (better error message) was dismissed as lipstick — the install still fails, just with nicer wording. This is the maximum fix that actually prevents the failure path by checking the resolved tag against a hardcoded floor BEFORE downloading. ## Changes `scripts/install.sh`: - New `MIN_ATTESTED_VERSION="v0.18.0"` constant near the top - New `version_ge` helper using `sort -V` (handles v0.9.0 vs v0.10.0) - Moved three-layer verification resolution (config → env → flag) to before the download so $verify_attestation is known in time to gate network work - New pre-flight check: if verification is requested and the resolved tag is older than MIN_ATTESTED_VERSION, fail fast with a clean message listing recovery options (pin to newer version, --skip-attestation, or unset the env var / config). No binary download, no wasted SHA256 check. - Late `gh attestation verify` block now only handles the gh call itself — resolution and pre-flight moved upstream. `scripts/install.ps1`: - New `$minAttestedVersion = "v0.18.0"` constant - Pre-flight guard in the verification branch using PowerShell's [version] class for proper numeric comparison - Same error message content as install.sh `scripts/install.cmd`: - New `set "MIN_ATTESTED_VERSION=v0.18.0"` near REPO setup - Pre-flight guard shells out to PowerShell for semver comparison — Windows 10+ ships `powershell.exe` always, so no new runtime dependency. Hand-parsing semver in cmd was tried and rejected as too fragile for prerelease tags and non-numeric components. `apps/marketing/.../installation.md`, `apps/hook/README.md`, `README.md`, `scripts/install.sh --help`: - Replaced every user-facing `v0.17.1` example with `vX.Y.Z` placeholder. The placeholder pattern already exists in the "Verifying your install" section, so this is just consistency. - install.sh --help adds a link to the releases page so users know where to find actual tag values. `.agents/skills/release/SKILL.md`: - New Phase 4 checklist step: before shipping the first attested release, verify MIN_ATTESTED_VERSION in all three installers matches the tag being cut. The constant is bumped ONCE and never again — it's a permanent floor, not a moving target. If the first post-merge release is not v0.18.0, the skill updates the constant in the same commit as the version bump so the installers served from plannotator.ai activate the new floor at the same moment the first attested release becomes fetchable. `scripts/install.test.ts`: - New test asserts all three installers hardcode MIN_ATTESTED_VERSION, use appropriate version comparison for their dialect, and contain the "predates" error message - New test asserts install.sh and --help text no longer contain `v0.17.1` as a pinned example 38 install tests pass (was 36). Smoke-tested install.sh end-to-end: - `--version v0.17.1 --verify-attestation` → pre-flight rejects cleanly, no download attempted, exit 1 with actionable error - `--version v0.18.0 --verify-attestation` → pre-flight passes, script proceeds to download (404 as expected since v0.18.0 is not yet released) - `--version v0.17.1` (no verify) → pre-flight skipped, normal download path For provenance purposes, this commit was AI assisted. * fix(install): close PS injection + move Windows pre-flight before download PR #512 review cycle 8 raised three related findings, all verified against actual code. ## Critical: PowerShell command injection in install.cmd (Finding 2) Line 228 of the previous install.cmd passed the version comparison to PowerShell by interpolating delayed-expansion variables directly into the command string between single-quoted literals: for /f "delims=" %%i in ('powershell -NoProfile -Command "try { if ([version]'!TAG_NUM!' -ge [version]'!MIN_NUM!') { 'yes' } } catch {}"') do set "VERSION_OK=%%i" The arg parser rejected leading-dash values but not quotes or semicolons, so a user passing install.cmd --version "0.18.0'; calc; '0.18.0" produced the PowerShell command try { if ([version]'0.18.0'; calc; '0.18.0' -ge [version]'0.18.0') { 'yes' } } catch {} PowerShell permits statement sequences inside `if` condition parentheses — the last value is used — so `calc` executed as a side effect during the first evaluation phase. Attacker-controlled --version from a CI/CD wrapper (PR titles, external tag sources, etc.) equals arbitrary code execution as the invoking user. Fixed by passing the version strings via environment variables ($env:TAG_NUM, $env:MIN_NUM) instead of interpolating them into the PowerShell command string. PowerShell reads $env: values as raw strings and never parses them as code. The [version] cast throws on invalid input, catch {} swallows it, VERSION_OK stays empty, and the guard rejects — safe fail with a slightly less helpful but correct error message. ## Structural: Windows pre-flight ran post-download (Findings 1 & 3) install.sh was already restructured in the previous commit to run the three-layer resolution + MIN_ATTESTED_VERSION guard BEFORE the binary download, so users hit the "predates attestation support" error without wasting bandwidth. install.ps1 and install.cmd drifted — their resolution and pre-flight blocks stayed in their original post-SHA256 positions, meaning the binary was always downloaded and SHA256-verified even when the requested tag was doomed to fail provenance verification. The "Pre-flight: reject the verification request before downloading" comments were lies copied from install.sh. This commit moves both Windows installers' resolution + pre-flight blocks upstream of the download: install.ps1: resolution + pre-flight now run immediately after `Write-Host "Installing plannotator $latestTag..."`, before $tmpFile is created or Invoke-WebRequest runs. The late gh-call block keeps only the gh attestation verify call itself. install.cmd: same restructure. The late block keeps only the where-gh check and gh invocation. The `del "!TEMP_FILE!"` calls inside the rejection branch are gone (TEMP_FILE doesn't exist yet when the guard runs). ## Tests Added two new regression guards to scripts/install.test.ts: 1. Order-aware check for all three installers: the resolution block's opening line must appear textually BEFORE the curl / Invoke-WebRequest download line. Uses indexOf to compare positions. Catches any future regression that drifts the pre-flight back after download. 2. Injection-safe pattern check for install.cmd: asserts the PowerShell command references $env:TAG_NUM / $env:MIN_NUM and does NOT interpolate !TAG_NUM! / !MIN_NUM! between single quotes in any [version] cast. 40 install tests pass (was 38). Smoke-tested install.sh with --version v0.17.1 --verify-attestation — rejects cleanly with no download, same as before. For provenance purposes, this commit was AI assisted. * fix(install): close cycle-9 gaps — CI coverage, v-strip, prerelease handling PR #512 cycle 9 review surfaced three real findings, all verified. ## Finding 1 (important): Windows CI never exercised the attestation path The Windows integration job ran `install.cmd v0.17.1 --skip-attestation`, which bypasses every bit of logic this PR shipped: three-layer opt-in resolution, MIN_ATTESTED_VERSION pre-flight, $env:-based PowerShell version comparison, and the gh attestation verify call. A runtime bug in any of those paths would not be caught by CI. `--skip-attestation` was passed intentionally because v0.17.1 predates attestation support — running without it hits the pre-flight and rejects. But that's the point: the REJECTION path is a real, valid end state we can assert against. The previous test conflated "install should succeed" with "test should pass"; the fix is to assert the correct behavior for an old version. Added a new CI step that runs `install.cmd v0.17.1 --verify-attestation` via Start-Process with stderr redirection to a temp file, then asserts: - exit code != 0 (pre-flight rejected) - stderr contains "predates" (rejection came from our guard, not some other failure mode like a network error or gh missing) This exercises on a real cmd.exe: - setlocal enabledelayedexpansion parser under the guard - three-layer resolution reaching the CLI flag layer - the :~1 substring (instead of the previous :v= global substitution) - the pre-release tag detection (negative path for stable tags) - the PowerShell shell-out with $env:TAG_NUM / $env:MIN_NUM - the [version] -ge comparison returning false - the "predates" error message block Can't test the success path (valid attested release) until the first post-merge release exists. Tracked for follow-up. ## Finding 2 (nit): !TAG:v=! is a global substitution, not anchored cmd's delayed-expansion string-substitution syntax `!VAR:str=repl!` replaces every occurrence of `str` globally. For all current semver tags (vX.Y.Z) this happens to strip exactly one `v` by coincidence. A hypothetical future tag like v1.0.0-rev2 would become 1.0.0-re2, which [System.Version] can't parse, silently misclassifying the failure as "predates attestation support" (see Finding 3). install.ps1 line 121 uses `-replace '^v', ''` which is properly regex-anchored. install.cmd had no anchored equivalent. Fixed by using `!TAG:~1!` — substring from index 1 — which drops exactly the first character. Safe because TAG is guaranteed to start with `v` by the normalization step upstream (line ~141). ## Finding 3 (nit): Pre-release tags misdiagnosed on Windows [System.Version] doesn't support semver prerelease or build-metadata suffixes (e.g. v0.18.0-rc1). It throws on any `-` in the version string. The catch blocks in both Windows installers handled the throw but surfaced wrong/confusing errors: install.sh: handles prereleases correctly via `sort -V` (POSIX version sort is semver-aware) — no issue. install.ps1: caught and printed "Could not parse version tags for provenance check" — accurate but doesn't explain WHY. install.cmd: swallowed silently, VERSION_OK stayed empty, printed "predates attestation support" — actively wrong, the problem isn't the version's age. Fixed in both Windows installers by detecting `-` in the tag BEFORE attempting the [version] cast: install.ps1: `if ($latestTag -match '-')` → dedicated error install.cmd: `if not "!TAG_NUM!"=="!TAG_NUM:-=!"` (native substitution check, no subshell, no metacharacter risk) Both emit a clear "pre-release tags aren't currently supported for provenance verification on Windows" message pointing users at --skip-attestation or a stable tag. Windows has no built-in semver comparator; adopting one would require NuGet or a custom parser. Explicit rejection with honest diagnosis is the pragmatic choice. ## Tests Three new regression guards in install.test.ts: 1. `install.cmd strips leading v via substring, not global substitution` — asserts `!TAG:~1!` is present and `!TAG:v=!` is gone. 2. `both Windows installers reject pre-release tags with a dedicated error` — asserts both scripts contain the "Pre-release tags" error message and the appropriate detection pattern for their dialect. 3. The new test.yml CI step doubles as a runtime regression guard — any break in the cmd pre-flight path that no longer matches "predates" in stderr, or returns 0, fails CI. 42 install tests pass (was 40). Windows CI will now exercise the pre-flight rejection path end-to-end for the first time. For provenance purposes, this commit was AI assisted. * fix: cycle-10 review — split attest job, assert binary preservation, misc PR #512 cycle 10 raised four findings, all verified. ## Finding 1: id-token/attestations permissions granted to build on PRs The build job in release.yml had `id-token: write` and `attestations: write` at the job level with no conditional guard. On PR triggers, those permissions were live for every build step (checkout, bun install, bun build, compile) even though the attestation step itself was gated by `if: startsWith(github.ref, 'refs/tags/')`. Narrow-but-real attack surface: a trusted contributor's malicious PR injecting code into a build step could mint an OIDC token authenticating as the repo identity. Fork PRs are automatically protected (GitHub suppresses OIDC tokens on forks), but same-repo contributor compromise is a realistic risk in a project with external contributors. Fixed by splitting attestation into its own job: build: contents: read only. Runs on all triggers. Compiles binaries and uploads them as the `binaries` artifact. No OIDC capability anywhere in the job. attest: needs: build, if: tag push only. contents: read + id-token: write + attestations: write. Downloads the binaries artifact and runs attest-build-provenance. Permissions are only live when we're actually producing an attestation — never on PR dry-runs. release: needs: attest (was: needs: build). Still tag-only. The dependency chain guarantees the attestation exists in the GitHub attestation store before the release's binaries are published, closing the race window where a user could pull the binary and gh attestation verify would fail because the bundle hadn't propagated yet. npm-publish: unchanged. Still needs: build. Still has id-token: write for `npm publish --provenance`. The reviewer flagged only the build job; npm-publish's id-token grant is scoped to that one job and is actually used by the provenance flag. ## Finding 2: CI test promised a binary-preservation check but didn't do one The `Attestation pre-flight rejects v0.17.1` step contained a multi-line comment promising to verify the rejected run didn't overwrite the previously-installed binary. No assertion code followed — just a Write-Host success line. The test claimed more than it delivered. Added actual baseline capture + comparison: - Before running the rejection test, capture the binary's SHA256 and LastWriteTime from the prior Gemini-merge step. - After the rejection, recompute both and assert they match. - Any drift throws: catches future regressions that re-introduce the post-download pre-flight pattern (the pre-flight correctly rejects but only after downloading and overwriting the file). ## Finding 3: install.ps1 dead-code comment about flag precedence Line 111 read "-SkipAttestation beats -VerifyAttestation if both passed" but the upfront mutex guard (lines 13-16) exits 1 if both flags are present. The "beats" scenario is unreachable. The comment misleads a future reader into thinking the late ordering handles the mutual exclusion and is safe to remove the early guard — which would be backwards. Replaced with a comment that explicitly notes the mutex guard at the top of the script makes the two branches mutually exclusive by construction. ## Finding 4: install.sh `cd` inside `&&` condition leaked CWD on failure The skills-install block chained `git clone ... && cd ... && git sparse-checkout set ...`. If clone succeeded but sparse-checkout failed, the short-circuit skipped the `cd -` and `rm -rf "$skills_tmp"` later ran with the shell's CWD still inside the to-be-deleted directory. On Linux/macOS this silently "works" — the inode is unlinked but the process keeps its cwd reference — so nothing visibly breaks (all downstream code uses absolute paths). But it's structurally wrong: install.ps1 and install.cmd both use Push-Location/pushd for the same logic. Restructured to run the entire clone → sparse-checkout → verify → copy sequence inside a single `(...)` subshell, with `cd`s scoped to the subshell. The parent shell's CWD is unchanged regardless of which step fails, so the subsequent `rm -rf` always runs from a stable location. Any failure in the chain short-circuits to the else branch with a clean skip message. Also merged the two `[ -d ]` / `[ ls -A ]` guards into the chain so the "apps/skills empty" case is now reported in the skip message rather than being silently suppressed. 42 install tests pass. For provenance purposes, this commit was AI assisted. * fix(install): set MIN_ATTESTED_VERSION to v0.17.2, remove skill bump note Earlier cycles hardcoded MIN_ATTESTED_VERSION="v0.18.0" across the three installers as a best-guess for the first post-merge release, and I added a one-time bump instruction to the release skill as insurance in case the guess was wrong. The guess was wrong — the next release is v0.17.2 (patch bump, not a minor bump). Updated the constant in all three installers and the matching test assertions. No other version references in the shipped error messages need changing because they read MIN_ATTESTED_VERSION from the variable at runtime. Also removed the "⚠️ One-time MIN_ATTESTED_VERSION bump" section from .agents/skills/release/SKILL.md entirely. With the constant now set to the actual next release tag, there's nothing for the release agent to bump at release time — the constant is already correct. Baking a one-time action into a recurring release skill was the wrong place for it; every future release agent would read the warning, confirm it's already set, and move on. Noise in a workflow that's supposed to be tight. If the next release version ever differs from v0.17.2 (e.g. we decide to skip to v0.18.0 or go straight to v1.0.0), the PR cutting that release will need to update MIN_ATTESTED_VERSION in the three installers. That's an ad-hoc fix, not a recurring skill concern. Smoke test with the new value: - install.sh --version v0.17.1 --verify-attestation → rejects with "first attested release is v0.17.2" - install.sh --version v0.17.2 --verify-attestation → passes pre-flight, proceeds to download (404 as expected since v0.17.2 is not yet released) 42 install tests pass. For provenance purposes, this commit was AI assisted. * feat(release): ship native ARM64 Windows binaries Bun v1.3.10 (February 2025) promoted bun-windows-arm64 from preview to a stable cross-compile target, which makes native ARM64 Windows builds a 15-line change instead of a project. Adopted immediately so ARM64 Windows users get native-speed binaries instead of the x86-64 emulation tax. Earlier cycles of this PR shipped two temporary workarounds for the absence of a native ARM64 binary: - install.ps1 detected ARM64 and fell back to $arch="x64" with a Write-Host notice that the user was running via emulation. - install.cmd hardcoded PLATFORM=win32-x64 and let ARM64 hosts pass the arch check without differentiation. Both are now obsolete and have been replaced with real architecture detection that selects the native binary. ## Changes release.yml: Added `bun-windows-arm64` to the compile matrix for both apps/hook/server/index.ts and apps/paste-service/targets/bun.ts. Output files are plannotator-win32-arm64.exe and plannotator-paste-win32-arm64.exe with matching .sha256 sidecars. Upload-artifact already globs `plannotator-*` so no change there. release.yml attest step: Added the two new ARM64 binaries to subject-path so they're covered by the SLSA build provenance attestation alongside the x64 builds. Both binaries sign with the same Sigstore bundle as the rest of the matrix. install.ps1: Restored the proper ARM64 detection that the earlier fallback replaced. On 64-bit Windows, $arch is "arm64" when PROCESSOR_ARCHITECTURE equals "ARM64", otherwise "x64". The emulation-fallback Write-Host notice is gone — users now get native binaries and don't need to be told about emulation. install.cmd: Replaced the unconditional `set "PLATFORM=win32-x64"` with a set of conditional assignments keyed off PROCESSOR_ARCHITECTURE and PROCESSOR_ARCHITEW6432 (the latter covers the edge case of a 32-bit tool launching install.cmd on an ARM64 machine via WoW64). PLATFORM is left empty if neither variable indicates AMD64 or ARM64, which triggers the "does not support 32-bit Windows" error path. The :arch_valid label and its gotos are gone — the new logic is linear and doesn't need a label. install.test.ts: Updated the install.ps1 ARM64 test to assert the native arm64 branch (no more "runs via emulation" text) and added a new install.cmd test verifying both PLATFORM branches are present. 43 install tests pass (was 42). ## CI coverage caveat windows-latest is x86-64, so the Windows integration job still exercises install.cmd against the x64 binary path. ARM64 has no CI runner coverage yet — we're shipping ARM64 binaries on trust that Bun's cross-compile produces working executables. That's the same trust we extend to linux-arm64 builds (also x-compiled from an ubuntu-latest runner). GitHub Actions does offer a windows-11-arm runner that could be added later; tracked as follow-up since it has availability and pricing implications. Closes #517. For provenance purposes, this commit was AI assisted. * fix(install.ps1): detect ARM64 host through WoW64 too, matching install.cmd Self-review catch on top of the ARM64 support commit. My install.ps1 architecture detection only checked \$env:PROCESSOR_ARCHITECTURE, which reports the architecture the CURRENT PowerShell process is running under — not the host architecture. On ARM64 Windows, a 32-bit PowerShell process (rare, but possible) would see PROCESSOR_ARCHITECTURE=X86, miss the "ARM64" branch, fall through to \$arch = "x64", and download the emulated x64 binary instead of the new native arm64 build. install.cmd already handles this correctly via PROCESSOR_ARCHITEW6432, which is set only in 32-bit WoW64 processes and holds the host architecture. install.ps1 was the odd one out. Fixed by checking PROCESSOR_ARCHITEW6432 first and falling back to PROCESSOR_ARCHITECTURE. Now both Windows installers follow the same detection logic regardless of process bitness. Also added an explicit error branch for unrecognized architectures (anything that isn't AMD64 or ARM64) instead of silently assuming x64. Test updated to assert both env vars are referenced. For provenance purposes, this commit was AI assisted. * fix(install): cycle-12 review — consistency test, dead code, finally, docs Four findings addressed. Two findings rejected. ## Finding 1 (nit): MIN_ATTESTED_VERSION triplicated without CI consistency Added a cross-file consistency test in install.test.ts that extracts the version literal from each of install.sh, install.ps1, install.cmd via regex and asserts all three match. A future bump that updates only one or two files now fails CI loudly. The per-file tests still exist (they check each file contains the current literal), but the new test catches drift where each file is internally consistent with itself but differs from the others. ## Finding 4 (nit): Write-Error + exit 1 dead code in install.ps1 Verified against the actual file: $ErrorActionPreference = "Stop" is set at line 8 and never modified. All six Write-Error sites are dead- end paths — five outside any try/catch, one inside a catch block (line 147) where Write-Error raises a new terminating error that propagates past the catch and exits the script with code 1 (PowerShell default). The `exit 1` lines that followed were never reachable. Dropped the six unreachable `exit 1` lines. Added a comment at the first occurrence explaining the Stop + Write-Error semantics so future maintainers don't re-add them. Behavior is unchanged at runtime — every error path still exits with code 1 via PowerShell's default unhandled-terminating-error handling. ## Finding 5 (nit): Pop-Location not in finally block Verified the reviewer's claim in install.ps1 lines 384-403. The skills install wraps git clone, Push-Location, and Copy-Item calls in a single try block, with Pop-Location on the success path. If Copy-Item throws under ErrorActionPreference=Stop, catch runs without popping, and the subsequent Remove-Item deletes a directory the PowerShell location stack still points into. A naive `finally { Pop-Location }` would introduce a new bug: Pop-Location throws on an empty stack, which happens when git clone silently fails and Push-Location is never reached. Used a nested-try pattern instead: try { git clone ... # native, no throw if (Test-Path "$skillsTmp\repo") { # guard against clone failure Push-Location "$skillsTmp\repo" try { ...operations... } finally { Pop-Location # always runs IF pushed } } } catch { Write-Host "Skipping..." } Traced all four failure modes: - clone fails silently → repo dir missing → skip inner block → no push, no pop → clean exit - clone succeeds → push succeeds → operations fail → finally pops → outer catch fires - clone + push + operations all succeed → finally pops cleanly - push itself throws (permissions) → outer catch fires, nothing to pop ## Finding 6 (P1): CMD/ps1 ARM64 breaks pinned pre-v0.17.2 tags The original plan was a runtime x64-fallback on 404, but the simpler product-level framing is: v0.17.2 is the first fully-supported version for pinning. Pre-v0.17.2 tags predate native ARM64 Windows (no win32-arm64 asset exists) and predate attestation support (pre-flight rejects). Users pinning to older tags are outside the supported matrix; the failure modes are explicit (404 / clean rejection), not silent corruption. Documented in the three install docs: - apps/marketing/.../installation.md: full "Supported versions" paragraph explaining the floor, what fails, and recovery paths - README.md: one-line note folded into the existing provenance sentence ("Version pinning, native ARM64 Windows, and SLSA provenance are supported from v0.17.2 onwards — see installation docs for details") - apps/hook/README.md: same tight one-liner pattern README.md and apps/hook/README.md stay bloat-free; the canonical explanation lives in the marketing docs. ## Findings rejected - **Finding 2 (P3, sort -V misorders prereleases):** plannotator doesn't ship prerelease tags. A user pinning to a hypothetical vX.Y.Z-rc1 would 404 at the download step before the sort -V misordering matters. Moot in practice. - **Finding 3 (important, sort -V is GNU-only):** FALSE POSITIVE. Tested on macOS 26.3.1 running sort 2.3-Apple (197) — both -V and --version-sort are supported and work correctly, including for prerelease suffixes. Apple forked BSD sort and added -V years ago. The reviewer's claim cites outdated reference material about historical BSD sort. 44 install tests pass (was 43). For provenance purposes, this commit was AI assisted. * test: anchor MIN_ATTESTED_VERSION consistency regexes to line start Self-review catch: the cross-file consistency test added in the prior commit matched the assignment form anywhere in each file. No current comment triggers a false positive, but a future comment like `# Example: MIN_ATTESTED_VERSION="v0.17.0"` would match first and shadow the real assignment, causing the test to report the wrong value or pass when it shouldn't. Hardened by adding /m flag and ^ anchor. The real assignments in all three installers are flush-left at the top of their files, so requiring line-start is both safe (won't reject current code) and stricter (future comments with leading whitespace or other prefixes are ignored). 44 tests still pass. For provenance purposes, this commit was AI assisted. * fix: cycle-13 review — checksum cleanup leak + Gemini CI coverage Two findings addressed. Two pre-existing findings flagged but not in scope. ## Finding 3 (nit): CHECKSUM_FILE leak on download failure install.cmd's checksum download error path deleted TEMP_FILE but omitted CHECKSUM_FILE. curl -o creates the output file before it receives data, so a network failure or HTTP error leaves a 0-byte or partial file in %TEMP% that the script never cleans up. The symmetric cleanup for TEMP_FILE elsewhere in the script makes this an accidental omission, not an intentional design choice. Added `if exist "!CHECKSUM_FILE!" del "!CHECKSUM_FILE!"` inside the error block, matching the existing cleanup discipline. ## Finding 1 (nit): Windows CI readback misses Gemini .toml files The `Verify Claude Code slash command files contain the shell- invocation prefix` step in the Windows integration job verified the three `.md` files at %USERPROFILE%\.claude\commands\ but not the two `.toml` files at %USERPROFILE%\.gemini\commands\. Both sets of files use the `^^!` cmd escape pattern that this PR added, and a future regression that drops a `^` from the Gemini echoes would slip past CI even though install.test.ts catches it statically. Extended the readback step to also verify plannotator-review.toml and plannotator-annotate.toml contain the `!{plannotator ...}` invocation form. Same regression class, same guard, same runner — the earlier Gemini-merge fixture step already seeds ~/.gemini/settings.json, which causes install.cmd's Gemini block to fire and write the .toml files alongside the Claude Code ones, so no additional setup is required. ## Findings rejected (out of scope, pre-existing) - **install.cmd vs install.ps1 install location divergence:** cmd installs to %USERPROFILE%\.local\bin while ps1 installs to %LOCALAPPDATA%\plannotator. A user who switches between the two Windows installers ends up with hooks.json pointing at one location and an orphan binary at the other. Pre-existing structural divergence, requires picking a canonical location and migrating users on whichever installer changes. Out of scope for this PR. - **install.sh Gemini merge throws on user's malformed JSON:** if ~/.gemini/settings.json is invalid JSON, the embedded `node -e` call exits non-zero, set -e propagates, and the install aborts mid-run after the binary is in place but before slash commands are written. Pre-existing — the Gemini block predates this PR. Worth a follow-up but not in scope here. 44 install tests pass. For provenance purposes, this commit was AI assisted. * docs: update stale v0.17.1 references in script comments to vX.Y.Z Two cosmetic comment fixes flagged during the cycle-13 self-review. The user-facing examples and docs were updated to vX.Y.Z in cycle 5, but two inline code comments still referenced the old concrete version: - scripts/install.sh:129 — "Positional form: install.sh v0.17.1 (matches install.cmd interface)" - scripts/install.cmd:71 — "Positional form: install.cmd v0.17.1 (legacy interface)" Both updated to vX.Y.Z so the in-code comments match the rest of the documentation. No behavior change. 44 install tests pass. For provenance purposes, this commit was AI assisted. * docs(skill): update release skill platform/binary counts for ARM64 Windows Two stale references in .agents/skills/release/SKILL.md after the ARM64 Windows binaries were added: - "5 platforms (macOS ARM64/x64, Linux x64/ARM64, Windows x64)" → "6 platforms (macOS ARM64/x64, Linux x64/ARM64, Windows x64/ARM64)" - "Compiles paste service binaries (same 5 platforms)" → "Compiles paste service binaries (same 6 platforms)" - "Generates SLSA build provenance attestations for all 10 binaries" → "Generates SLSA build provenance attestations for all 12 binaries" The first two predate this PR; the third was added in the cycle-1 commit and not bumped when the ARM64 Windows targets landed in the ARM64 commit. All three corrected together so the release agent sees an internally-consistent description of what the pipeline actually does. Verified against release.yml — 12 entries in the attest job's subject-path list, 12 compile commands in the build job, all six platforms (macOS arm64/x64, Linux x64/arm64, Windows x64/arm64) for both plannotator and paste-service. For provenance purposes, this commit was AI assisted.	2026-04-07 20:35:16 -07:00
Michael Ramos	7e8f914904	feat(plan): unified header dropdown + Agent Instructions (#515) ... * feat(plan): unify header settings into combined dropdown Replace the desktop sprawl (ModeToggle, Settings button, Export sub-dropdown) and the separate MobileMenu with a single PlanHeaderMenu modeled on ReviewHeaderMenu — one ActionMenu rendered at all breakpoints. Move version + release notes into the dropdown footer, drop the origin/agent badge, and stub an "Agent Instructions" item that copies the current plan plus an /api/external-annotations URL for external agents (real protocol body lands in a follow-up). Also fix the annotate-folder Send button: it was hidden whenever a file was opened via linkedDoc, so users in `plannotator annotate ./` couldn't submit feedback. Gate now allows linked-doc state through in annotate mode. Settings is always mounted instead of skipping when a linked doc is active, so the dropdown's Settings item works in every state. For provenance purposes, this commit was AI assisted. * feat(plan): real Agent Instructions clipboard payload Replace the placeholder string in handleCopyAgentInstructions with a real, contract-accurate markdown payload that teaches an external agent (Claude Code, Codex, custom scripts) how to: - read the current plan via GET /api/plan - POST single or batch annotations to /api/external-annotations - choose COMMENT vs DELETION vs GLOBAL_COMMENT and use originalText for inline highlighting (or skip it for sidebar-only) - list, delete-by-id, and delete-by-source for cleanup before reposting The payload lives in packages/ui/utils/planAgentInstructions.ts as buildPlanAgentInstructions(origin) — a single pure function that takes the base URL and returns the markdown. Plan and code-review modes have different annotation shapes, so each owns its own instructions module; the review counterpart will live alongside this file when it's added. For provenance purposes, this commit was AI assisted. * docs: list planAgentInstructions.ts in AGENTS.md utils enumeration For provenance purposes, this commit was AI assisted. * fix(plan): restore 'system' theme option in header dropdown The unified header dropdown only exposed light/dark, dropping the 'system' (follow OS) option that the previous ModeToggle and MobileMenu both supported. Clicking either explicit mode would silently overwrite a prior 'system' setting. Add it back as a third segmented control, matching the old MobileMenu's three-way layout. For provenance purposes, this commit was AI assisted. * refactor(ui): extract shared theme-mode icons PlanHeaderMenu and ThemeTab each inlined their own Sun/Moon/System SVGs (and the deleted MobileMenu had its own copy too). Pull them into a shared icons/themeIcons.tsx module so the iconography stays consistent across the dropdown and the settings tab. Use the cleaner Heroicons paths from ThemeTab as the canonical art. ReviewHeaderMenu still inlines its own copies — left alone for this PR since the broader review/plan menu unification is a separate follow-up. For provenance purposes, this commit was AI assisted. * fix(plan): use ReviewAgentsIcon for Agent Instructions menu item Reuse the magnifying-glass-with-cog icon already used for the Review Agents tab so the same visual marker identifies agent-related affordances across plan and review modes. For provenance purposes, this commit was AI assisted. * fix(plan): drop DELETION + unanchored COMMENT from agent instructions Plannotator's existing render and export pipeline only handles two shapes well: an anchored COMMENT with originalText, and a GLOBAL_COMMENT with no anchor. The previous instructions also documented: - COMMENT without originalText — sidebar shows an empty quote bubble ("") and the export header reads `Feedback on: ""`. The text survives but the framing is broken. - DELETION with explanatory text — the text is invisible in the sidebar and silently dropped on export, replaced by a generic template. Real information loss. Rather than expand the export pipeline to honor those edge cases, narrow the documented surface to the two shapes that work today: inline COMMENT (requires originalText) and GLOBAL_COMMENT (no anchor). DELETION is omitted entirely; agents that need to suggest a removal can post an inline COMMENT pointing at the phrase. For provenance purposes, this commit was AI assisted. * fix(plan): clear current new-settings hints, surface pulse on menu trigger The new-settings indicator infrastructure (hasNewSettings / version constant / markNewSettingsSeen) stays — it's the scaffolding for future hint flagging — but every current marker is removed: - Drop the four inline 'new' badges in Settings.tsx (Plan Width row, Quick Labels row, mobile + desktop tab navigators). - Drop the dead pulse on the standalone gear button — that button now lives inside <div className="hidden"> after the header refactor, so the pulse was unreachable anyway. Surface the pulse on the new PlanHeaderMenu trigger instead. App.tsx seeds local hasNewSettingsHints state from hasNewSettings(), passes it to the menu, and clears it (plus calls markNewSettingsSeen) when the user opens Settings from the dropdown. For provenance purposes, this commit was AI assisted. * docs: list icons/ subdirectory in AGENTS.md components tree For provenance purposes, this commit was AI assisted. * fix(server): require originalText for plan COMMENT annotations The plan-mode external-annotation contract documents that COMMENT annotations must carry an originalText anchor; sidebar-only feedback uses GLOBAL_COMMENT instead. transformPlanInput already enforced this for DELETION but silently defaulted COMMENT's originalText to "" when missing — producing a broken render (empty quote bubble in the sidebar, exported as `Feedback on: ""`). Reject COMMENT without a non-empty originalText and steer the caller toward GLOBAL_COMMENT in the error message so the fix is self-evident to misbehaving agents. For provenance purposes, this commit was AI assisted. * chore: drop dead new-hints scaffolding leftovers After clearing the four current inline hint markers, two pieces of support code were unreachable: - The destructured setShowNewHints setter in Settings.tsx — the state itself stays as a stable mount-time snapshot for any future hint markers, but the setter has no callers. Drop it and add a comment explaining why the useState wrapper is intentional scaffolding. - The settings-ping keyframe in editor/index.css — only the deleted gear-button pulse referenced it; the new pulse on PlanHeaderMenu's trigger uses Tailwind's built-in animate-ping instead. For provenance purposes, this commit was AI assisted.	2026-04-07 15:03:30 -07:00
Michael Ramos	b375a804b2	feat(review): AI review agents, local worktree, and UI polish (#491) ... * feat(review): add Codex AI review agent with live logs, --local worktree, and panel UI Hook up Codex as the first AI review agent in the code review system: - Spawn `codex exec` with Codex's native review prompt and output schema - Parse structured findings (ReviewOutputEvent) and push as external annotations - Annotations appear inline in the diff viewer pinned to specific lines - Review verdict (correct/incorrect + confidence + explanation) displayed in panel Agent job infrastructure enhancements: - Server-side command building via `buildCommand` callback (providers don't need frontend commands) - Result ingestion via `onJobComplete` callback (reads output file, transforms findings) - `addAnnotations` method on external annotation handler (bypasses HTTP for server-internal producers) - Live stderr streaming via `job:log` SSE events with 200ms buffer-and-flush - `cwd` and `summary` fields on AgentJobInfo PR review with --local worktree: - `plannotator review <PR_URL> --local` creates a temp git worktree with the PR branch - Agent gets full local file access without touching the user's working tree - Hybrid server mode: both prMetadata (platform features) and gitContext (local access) - Automatic worktree cleanup on session end - Runtime-agnostic worktree primitives in packages/shared/worktree.ts Panel UI redesign: - Findings | Logs tab system with underline-style tabs - LiveLogViewer component with auto-scroll, truncation, and copy - Review verdict card (correct/incorrect with confidence and explanation) - Pending state with labeled "Review Verdict — Pending..." (not skeleton bars) - Job card click opens detail panel directly (removed separate icon button) - Dismissed annotation tracking (deleted annotations persist as "dismissed" in panel) - Copy all annotations as formatted markdown - Worktree badge in header with info dialog showing path - CopyButton extended with inline variant for reuse - ConfirmDialog extended with wide option For provenance purposes, this commit was AI assisted. * style(review): UX polish pass — visual quality improvements across code review UI - VerdictCard: remove AI-template left-border, use background-only tint - Inline annotations: 6px radius, subtle shadow, hover elevation, action button scale - File tree: tighter indentation (4 + depth*10), reduced container padding - Select dropdowns: normalized to 4px border-radius matching pierre diffs - Dockview tabs: close button pushed to far right with margin-left auto, visible at 0.25 opacity - PR icons moved from sidebar to header (next to PR link) - AnnotationRow: translate-x hover feedback - Border-radius normalized to `rounded` (4px) across all components - Type scale consolidated: text-[8px]/[9px] → text-[10px], text-[11px] → text-xs - Sidebar tab hit targets increased (px-2.5 py-1.5, w-4 icons) - Sticky file group headers in annotation sidebar - FileTree controls collapsed (worktree + diff selectors share one row) - Tab micro-animations (transition-all duration-150) - ScrollFade component for gradient indicators on scrollable containers - Prose containment: max-w-2xl + px-6 padding on PR Summary/Comments/Checks panels - MarkdownBody: leading-relaxed for comfortable reading - PR Comments: surface lift (bg-muted/10), hover feedback, author font-semibold - PR Checks: link affordance (text-primary, underline on hover, external link icon) For provenance purposes, this commit was AI assisted. * feat(review): PR comments panel — search, filter, collapse, navigation, and polish Comments panel enhancements: - Search: real-time text filter across author and body, match count display - Keyboard navigation: j/k to move between comments, scroll-to-selected - Sort: toggle between oldest/newest first - Collapsible comments: click header to collapse, collapse/expand all controls - Author exclusion filter: click authors to hide their comments (not inclusion) - Comment actions: hover-reveal "View on GitHub" link + copy button (bottom-right) - Review URLs: PRReview type now includes optional url field, populated from GitHub API PR Summary fixes: - Label contrast: use theme foreground color for label text instead of raw GitHub hex - Linked issues: replaced broken hardcoded SVG with proper GitHub Octicons issue-opened icon Data plumbing: - platformUser exposed through ReviewStateContext for "Mine" filtering - Panel wrapper changed to overflow-hidden for sticky toolbar support - "Commented" review badge hidden (noise — only show Approved/Changes Requested/Dismissed) For provenance purposes, this commit was AI assisted. * feat(review): inline review threads with outdated/resolved state and diff hunk previews PR Review Threads: - Fetch inline code review comments via GitHub GraphQL (reviewThreads query) - PRReviewThread and PRThreadComment types with isResolved, isOutdated, path, line, diffSide - ThreadCard component: file/line context, Outdated/Resolved badges, nested replies - Resolved/outdated threads: gradient fade on body with "Show full comment" expand - GitLab: reviewThreads placeholder (TODO: parse DiffNote positions from notes) DiffHunkPreview: - Renders diff hunks using @pierre/diffs FileDiff component (read-only, compact) - Full theme integration: reads computed CSS vars, injects via unsafeCSS (same as main DiffViewer) - Respects user font settings from ReviewState context - Handles bare GitHub diffHunk format (prepends synthetic file headers for pierre parsing) Comments Panel Polish: - Comment cards: bg-card + subtle shadow for depth and isolation from panel background - Hover: shadow elevation (0_2px_6px) for interactive feedback - Thread cards: dimmed shadow for resolved/outdated, full shadow for active - Prose padding: px-8 (32px) across all dockview panels (Summary, Comments, Checks, Findings) For provenance purposes, this commit was AI assisted. * feat(review): Claude Code agent, cross-repo --local, render fixes Claude Code review agent: - claude-review.ts: prompt (adapted from code-review plugin), command builder (dontAsk + granular allowedTools/disallowedTools), JSONL stream output parser - Prompt sent via stdin (not argv) to avoid quoting/variadic flag conflicts - stream-json --verbose for live JSONL streaming + final structured_output - Same schema as Codex — transformReviewFindings is now provider-agnostic Agent jobs infrastructure: - stdout capture (captureStdout option) for providers that return results on stdout - stdin prompt writing (stdinPrompt option) for providers that read prompt from stdin - cwd override in buildCommand return for providers without -C flag - await stdoutDone before onJobComplete to prevent drain race condition - job.prompt field for transparent prompt display in detail panel Cross-repo --local: - Detect same-repo vs cross-repo via parseRemoteUrl comparison - Cross-repo: shallow clone via gh/glab repo clone (--depth 1 --no-checkout + targeted fetch) - Cross-repo uses platform diff (gh pr diff) for display, clone for agent file access - Same-repo: existing worktree path unchanged - Cleanup: rmSync for clones, worktree remove for same-repo Performance: jobLogs context split - Separate JobLogsContext to prevent high-frequency log SSE from re-rendering all panels - Only ReviewAgentJobDetailPanel subscribes to JobLogsProvider - Standard React pattern: split contexts by update frequency Image error handling: - SafeHtmlBlock component wraps dangerouslySetInnerHTML with img onerror handlers - Broken images (expired GitHub JWTs) hide on first 404 instead of flickering - Prevents console 404 flood from re-render retry loops For provenance purposes, this commit was AI assisted. * fix(review): decontaminate --local from diff pipeline, fix worktree setup, default local for PRs The --local flag was setting gitContext from the worktree, which contaminated the diff rendering pipeline — causing pierre "trailing context mismatch" errors because /api/file-content read worktree files instead of using the GitHub API. Root cause: gitContext serves two purposes — diff pipeline (file contents, diff switching, staging) and agent sandbox (cwd for agent processes). These are now properly separated via a new agentCwd option on ReviewServerOptions. Changes: - Add agentCwd to ReviewServerOptions, independent of gitContext - Agent handler (getCwd, buildCommand, onJobComplete) prefers agentCwd - Stop setting gitContext in --local PR path — diff pipeline untouched - Revert band-aid !isPRMode guards (no longer needed) - Fix same-repo worktree: fetch origin/<baseBranch> so agents see correct diff - Fix cross-repo clone: create local branch at baseSha for git diff accuracy - Fix FETCH_HEAD ordering: fetch base branch before PR head (createWorktree needs PR tip) - Fix macOS path mismatch: realpathSync(tmpdir()) so agent paths strip correctly - Change buildCodexReviewUserMessage signature from GitContext to focused options - Make --local the default for PR/MR reviews (--no-local to opt out) - Pass agentCwd to client for worktree badge display For provenance purposes, this commit was AI assisted. * style(review): unify panel headers, responsive buttons, dockview polish - Unify all panel headers at 33px via --panel-header-h CSS variable - FileHeader now uses shared variable instead of hardcoded 30px - Dockview tab bar height, font size, and padding aligned with sidebars - FileTree header uses fixed height instead of padding-based sizing - Consistent border opacity (border-border/50) across all panels - Consistent font weight (font-semibold) on all header labels - Remove dockview tab focus outline (::after pseudo-element) - Dockview tab close button pushed to right edge - Tab bar void area uses muted background - Top app header compacted from h-12 to py-1 - Sidebar footer: copy button and diff stats side by side - FeedbackButton responsive labels (Send/Post at md, full labels at lg) - Move ReviewAgentsIcon to packages/ui for shared use - Agent empty state uses shared ReviewAgentsIcon instead of hardcoded SVG - Sidebar header label truncates when narrow (tabs never clip) - Detail panel prompt disclosures get proper spacing - React.memo on PR tab components (PRSummaryTab, PRCommentsTab) - Inline onerror on img tags for broken GitHub image handling For provenance purposes, this commit was AI assisted. * fix: type assertion for Bun stdin FileSink Bun's proc.stdin is typed as `number | FileSink` but we need to call .write() and .end() on it. Cast to FileSink to satisfy tsc --noEmit. For provenance purposes, this commit was AI assisted. * fix(review): XSS in sanitizeHtml, git flag injection, cross-repo ref mismatch Security: - Remove `onerror` from DOMPurify ALLOWED_ATTR — was allowing arbitrary JS execution via PR descriptions containing `<img onerror="...">`. Replace with SafeHtml component that attaches error handlers via useEffect + ref. - Add `--` end-of-options separator to git fetch and git branch calls to prevent flag injection via crafted branch names from API responses. Bug fix: - Cross-repo clones now create both local branch AND remote-tracking ref (`refs/remotes/origin/<baseBranch>`) at baseSha, so agents can use either `git diff main...HEAD` or `git diff origin/main...HEAD`. Polish: - Add copy button to verdict card in agent detail panel - Remove hover:translate-x animation from finding rows - Reduce file tree indent per level, remove extra file offset - Add pr-action endpoint logging for debugging submit failures For provenance purposes, this commit was AI assisted. * chore: upgrade @pierre/diffs from 1.1.0-beta.19 to ^1.1.12 The beta pin was needed for processFile() API (expandable diff context), which shipped in 1.1.0 stable on March 14. We were 13 releases behind. Notable fixes in the upgrade path: - 1.1.5: Fix diffAcceptRejectHunk with partial FileDiffMetadata - 1.1.6: Patch parsing fix for renames and dotfiles - 1.1.8: Fix maxLineDiffLength regression May resolve intermittent "trailing context mismatch" errors in diff rendering. For provenance purposes, this commit was AI assisted. * fix(review): remove shell provider, flag injection, Claude log formatting, copy UX Security: - Remove shell provider from agent capabilities (unauthenticated RCE vector) - Move `--` before prRepo in `gh repo clone` to prevent flag injection Features: - Wire up formatClaudeLogEvent — Claude live logs now show readable text instead of raw JSONL - Sidebar annotations: copy + delete buttons appear on hover (no overlap) - Agent finding rows: copy button on hover (progressive disclosure) - Verdict card: copy button pushed to the right - File tree: reduced indent per level, files aligned with folders Infra: - PR action endpoint logging for debugging submit failures For provenance purposes, this commit was AI assisted. * fix: validate repo identifier to prevent flag injection in gh repo clone The `--` separator in `gh repo clone` separates gh args from git args, not positional args from flags. Using `--` before prRepo would break the git flags. Instead, validate that the repo identifier doesn't start with `-` to prevent flag injection via crafted PR URLs. For provenance purposes, this commit was AI assisted. * feat(review): Claude-specific review model with severity, reasoning, and multi-agent prompt Claude review agent now has its own schema, prompt, and transform — separate from Codex's P0-P3 priority model. Each provider uses its natural review style. Schema changes: - Claude findings use severity (important/nit/pre_existing) instead of priority (0-3) - Flat structure: file, line, end_line instead of nested code_location - description (single field) instead of title + body - reasoning field captures the validation chain per finding - summary with counts instead of overall_correctness/confidence Prompt: Converges the open-source Claude Code review prompt with the remote review service model. 4 parallel agents (Bug+Regression at Opus, Security at Opus, Code Quality at Sonnet, Guideline Compliance at Haiku), validation step, deduplication, severity classification. CLAUDE.md and REVIEW.md awareness. UI: Severity markers (colored dots) on finding rows. Collapsible reasoning section via <details>. New optional severity/reasoning fields on CodeAnnotation and the external annotation store — backward compatible, only set by Claude. Transform: transformClaudeFindings normalizes Claude output into the shared annotation format. Codex path (transformReviewFindings) is completely untouched. For provenance purposes, this commit was AI assisted. * fix: use bg-amber-500 for nit severity dot (bg-warning may not be defined) For provenance purposes, this commit was AI assisted. * fix: security hardening, debug cleanup, deduplication, Pi mirroring, findings UX Security: - Add -- separator to ensureObjectAvailable git fetch (worktree.ts) - Validate baseBranch against path traversal (..) before git ref operations - Use process.once('exit') instead of process.on for worktree cleanup Debug: - Gate debugLog behind PLANNOTATOR_DEBUG env var (no more unconditional writes) - Remove PARSE_OUTPUT_RAW dump that logged full JSON output to disk Deduplication: - Extract toRelativePath to packages/server/path-utils.ts (was duplicated in codex-review.ts and claude-review.ts) Pi extension: - Remove shell provider from capabilities - Add buildCommand callback to AgentJobHandlerOptions - POST handler calls buildCommand for server-side command synthesis UI: - Findings sorted by severity (important → nit → pre_existing) - Severity legend under findings header (colored dots) - Reasoning always visible (not collapsible) — fixes click navigation bug where <details> captured click events and broke annotation linkage - Full finding text shown (removed line-clamp-2 truncation) - Sidebar annotation hover actions aligned to the right - DiffHunkPreview: cancel requestAnimationFrame on unmount For provenance purposes, this commit was AI assisted. * fix: move toRelativePath import to top of claude-review.ts For provenance purposes, this commit was AI assisted. * fix: same-repo detection compares host, platform-aware comment links, remove design docs - Same-repo detection now compares both owner/repo AND hostname from the git remote URL against prMetadata.host. Prevents false positives on GitHub Enterprise where different instances share org/repo names. - "View on GitHub" label in PR comments tab now shows "View on GitLab" for GitLab MR comments based on the comment URL. - Remove internal design docs (PR_LOCAL_WORKTREE.md, AGENT_LIVE_LOGS.md) that were development artifacts, not user-facing documentation. For provenance purposes, this commit was AI assisted. * fix(review): show severity markers and reasoning in inline diff annotations The severity and reasoning fields from Claude findings were only visible in the agent detail panel, not in the inline diff annotations. Now: - DiffAnnotationMetadata carries severity and reasoning fields - DiffViewer passes them through when mapping annotations - InlineAnnotation renders colored severity dot and reasoning text For provenance purposes, this commit was AI assisted. * fix: prefix Claude findings text with [severity] tag Findings now show as "[important] description", "[nit] description", "[pre_existing] description" — consistent with Codex's [P0]/[P1] tags. For provenance purposes, this commit was AI assisted. * feat(pi): full agent review mirroring — stdin, stdout, live logs, result ingestion Pi extension agent-jobs handler now mirrors the Bun server's full capabilities: - stdin piping for Claude prompt delivery - stdout capture for Claude JSONL stream parsing - Live stderr streaming with 200ms buffer-and-flush for job:log events - Claude JSONL formatting via vendored formatClaudeLogEvent - onJobComplete callback for result parsing and annotation push - Full buildCommand integration in POST handler - jobOutputPaths tracking with cleanup on kill Pi serverReview.ts now wires buildCommand and onJobComplete with the same logic as the Bun review server — Codex and Claude commands are built server-side, results are parsed and transformed into external annotations. Runtime compatibility: replaced Bun.file/Bun.write in codex-review.ts with node:fs/promises equivalents (writeFile, readFile, existsSync) that work on both Bun and Node. Verified Bun build passes. Vendoring: vendor.sh now copies codex-review.ts, claude-review.ts, and path-utils.ts from packages/server/ with import path rewriting for the generated/ layout. Also: Review Prompt label, px-8 padding on agent detail header/tabs/logs. For provenance purposes, this commit was AI assisted. * fix: remove duplicate isPRMode declaration in Pi serverReview.ts For provenance purposes, this commit was AI assisted. * fix: include reasoning in all copy and feedback export paths - exportReviewFeedback: appends **Reasoning:** after finding text - Per-finding copy button: appends reasoning to copy text - Sidebar annotation copy: appends reasoning to copy text This ensures reasoning flows through Copy All, Send Feedback, and individual copy actions — not just the visual rendering. For provenance purposes, this commit was AI assisted. * fix: six verified findings — navigation, dedup, cleanup, copy, diff match, Windows paths 1. openDiffFile: clicking a finding now navigates to the correct file before selecting the annotation (was silently selecting in wrong file) 2. SEVERITY_STYLES: extracted to packages/ui/types.ts as shared constant, imported in both ReviewAgentJobDetailPanel and InlineAnnotation (was duplicated with per-render rebuild in InlineAnnotation) 3. killJob: added jobOutputPaths.delete calls to match Pi's version (was leaking two strings per killed job) 4. CommentActions: replaced hand-rolled copy with CopyButton inline variant (was reimplementing useState/clipboard/setTimeout pattern) 5. Branch mode prompt: changed from three-dot to two-dot to match the UI's actual diff computation (agent was reviewing different diff) 6. toRelativePath: uses path.relative + forward-slash normalization for Windows compatibility (was string prefix matching with / only) For provenance purposes, this commit was AI assisted. * perf: wrap ReviewSidebar in React.memo to prevent re-renders during log streaming Every job:log SSE event triggers setJobLogs in useAgentJobs, which re-renders App.tsx. Without memo, the sidebar re-renders on every event (~5/sec) even though its props (agentJobs.jobs, capabilities, callbacks) haven't changed. This caused visible flickering when a review tab was open during agent runs. React.memo shallow-compares props — all sidebar props are stable references (jobs array only changes on status events, callbacks are useCallback-wrapped), so the sidebar correctly skips re-renders during log streaming. For provenance purposes, this commit was AI assisted. * fix(security): remove find/ls/cat from Claude allowed tools, add glab CLI Security: Bash(find:*) allowed find -exec to spawn arbitrary subprocesses that bypassed --disallowedTools. Removed find, ls, and cat — Claude has Glob, Read, and Grep built-in which cover file access without shell exec. Feature: Added glab mr view/diff/list and glab api to allowed tools so Claude can inspect GitLab MR context in remote-mode reviews. For provenance purposes, this commit was AI assisted. * fix: Pi addAnnotations, Pi stdout drain, cross-repo exit codes Pi extension: - Add addAnnotations() to external-annotations.ts return object — serverReview.ts calls it when agent jobs complete but the method was missing (build/runtime error) - Change proc.on('exit') to proc.on('close') in agent-jobs.ts — Node's 'exit' fires before stdio streams drain, so stdoutBuf could be incomplete when onJobComplete parses Claude's JSONL result Cross-repo --local: - Check git checkout FETCH_HEAD exit code — throw if it fails so the outer catch falls back to remote-only with a clear warning - Log warning if baseSha fetch fails (non-fatal, agents just can't diff locally) For provenance purposes, this commit was AI assisted. * fix: FETCH_HEAD ordering, Pi worktree-aware cwd, SEVERITY_ORDER hoisted Critical: - Move ensureObjectAvailable before PR head fetch — it can overwrite FETCH_HEAD if baseSha needs fetching, causing createWorktree to check out the base commit instead of the PR head - Pi serverReview.ts: extract resolveAgentCwd() helper used by getCwd, buildCommand, and onJobComplete — was bypassing worktree-aware path resolution, causing agents to run in wrong directory Cleanup: - Hoist SEVERITY_ORDER to module scope in ReviewAgentJobDetailPanel (was recreated inside component body on every render) For provenance purposes, this commit was AI assisted. * fix: Bun/Pi parity — provider default, annotation error logging - Bun agent-jobs: change provider default from "shell" to "" (shell was removed from capabilities, default should match Pi) - Pi serverReview: log errors from addAnnotations in onJobComplete (Bun logs them, Pi was silently ignoring) For provenance purposes, this commit was AI assisted. * docs: add AI Code Review Agents guide with full prompt transparency New docs page covering: - Overview of Codex and Claude review agents - How findings work (severity/priority, reasoning, navigation) - Local worktree behavior (same-repo vs cross-repo) - Full transparency section with: - Claude multi-agent pipeline prompt (all 6 steps) - Claude command and allowed/blocked tools - Codex review prompt and command - Both output schemas (Claude severity + Codex priority) - Security notes (read-only, no network, local execution, no commenting) - Customization via CLAUDE.md and REVIEW.md For provenance purposes, this commit was AI assisted. * docs: add provenance links for Claude and Codex review integrations Credit Anthropic's Claude Code Review service, the open-source code-review plugin, and OpenAI's Codex CLI as the foundations for our review agent integrations. For provenance purposes, this commit was AI assisted. * fix: temp clone leak, j/k key conflict, thread header null line - Cross-repo: clean up localPath in catch block when fetch/checkout fails after clone succeeds (directory was leaking in /tmp) - Remove j/k/arrow keyboard navigation from PR comments panel — these shortcuts belong to the file tree only, both registering global handlers caused double-navigation - Thread header null guard: check thread.line before building range label to prevent "L12–null" for outdated GitHub threads For provenance purposes, this commit was AI assisted. * fix: hoist localPath for catch-block scope, validate baseSha format The previous rmSync(localPath) in the catch block was dead code — const declarations inside try are not in scope in catch (separate lexical environments per ECMAScript spec). The ReferenceError was silently swallowed by the inner try/catch, so temp directories still leaked on failed fetch/checkout. Fix: hoist `let localPath` before the try block so it's accessible in catch. Guard with `if (localPath)` since the error could occur before assignment. Also: validate baseSha is a hex SHA (40-64 chars) to prevent git flag injection via crafted API responses. Validate baseBranch rejects both '..' and '-' prefixes. For provenance purposes, this commit was AI assisted. * docs: rewrite AI Code Review guide for clarity and readability Restructured for a technical audience: shorter paragraphs, cleaner tables, removed em-dashes and filler prose, tightened the pipeline diagram, streamlined security section into scannable single-line items. For provenance purposes, this commit was AI assisted. * docs: rewrite transparency section with exact prompts and commands Replaced summarized/paraphrased transparency section with the actual prompts, commands, schemas, and tool allowlists as they exist in the code. One short security note at the top, then raw content. For provenance purposes, this commit was AI assisted. * docs: add mini TOC to transparency section For provenance purposes, this commit was AI assisted. * fix: stdout drain hang, Codex verdict override, Claude parse logging, memo removal Critical: - Race stdoutDone against 2s timeout after proc.exited — prevents permanent job hang when Bun's ReadableStream doesn't close after process exit. The process is dead; 2s is a cleanup deadline. Bug fix: - Codex verdict: override to "Issues Found" when P0/P1 findings exist, regardless of the freeform overall_correctness string. Prevents green "Correct" badge when Codex says "mostly correct but has issues." P2/P3-only findings still trust Codex's verdict. Observability: - Log Claude parse failures with buffer size and last 200 bytes so we can diagnose empty-findings cases. Performance: - Remove React.memo from ReviewSidebar — was blocking legitimate re-renders (job status, findings) to prevent cosmetic log flickering. The tradeoff was wrong. Docs: - Remove "shell" from CLAUDE.md capabilities table (provider was removed). For provenance purposes, this commit was AI assisted. * fix: type assertions for Bun ReadableStream async iteration Bun's proc.stdout/stderr support for-await at runtime but TypeScript's ReadableStream type doesn't declare [Symbol.asyncIterator]. Cast through unknown to AsyncIterable<Uint8Array> — standard Bun workaround, same pattern as the FileSink cast for stdin. For provenance purposes, this commit was AI assisted.	2026-04-06 12:15:27 -07:00
foxytanuki	26364543b2	fix(remote): support explicit local override (#481)	2026-04-04 13:17:16 -07:00
Michael Ramos	68418e2a35	feat(review): add dockview center workspace (#453) ... * chore(review): add dockview-react dependency * feat(review): add dockview center workspace * fix(review): stabilize diff search navigation * feat(review): add progressive file search UI * feat(review): streamline header controls * fix(review): portal annotation toolbar * refactor(review): streamline header and action menus * fix(review): make file header responsive * fix(review): align split diff drag handle * chore(review): hide review agents sidebar * fix(review): remount diff viewer on file switch * fix(review): wrap inline diff comments	2026-04-01 14:58:45 -07:00
Michael Ramos	d8a67beb1c	feat: agentic review — background job runner with SSE streaming (#443) ... * feat(server): add agent job manager with process lifecycle and SSE streaming Introduces the backend for agentic review — a server-agnostic job runner that spawns agent processes, monitors their lifecycle via proc.exited, and broadcasts status updates over a dedicated SSE stream. New files: - packages/shared/agent-jobs.ts: runtime-agnostic types, SSE helpers, state machine - packages/server/agent-jobs.ts: Bun HTTP adapter with spawn/kill/SSE/REST routes Wired into the review server with late-bound serverUrl, killAll on stop and process exit for defense-in-depth cleanup. Capability detection via Bun.which() for Claude Code, Codex CLI, and shell. For provenance purposes, this commit was AI assisted. * feat(ui): add Review Agents tab with job management and live status Adds the client-side UI for agentic review: - useAgentJobs hook: SSE primary with polling fallback, capabilities fetch - AgentsTab: shared component with provider dropdown, job cards, kill controls - ReviewAgentsIcon: magnifying glass with cog icon - ReviewPanel: new "Review Agents" tab conditional on detected capabilities - App.tsx: initializes hook and passes props to ReviewPanel Jobs show live elapsed time, status badges, annotation counts per source, and expandable error details. Stub command (sleep 10-30s) for initial testing. For provenance purposes, this commit was AI assisted. * fix(server): narrow proc.stderr type to satisfy TS Bun types proc.stderr as ReadableStream | number. Add typeof guard to narrow before passing to Response constructor. For provenance purposes, this commit was AI assisted. * fix(server): prevent stdout deadlock, stderr pipe stall, and worktree cwd mismatch - Change stdout from "pipe" to "ignore" — agents post findings via the annotations API, not stdout. Unconsumed pipe buffer caused deadlock. - Drain stderr continuously instead of reading after exit — prevents pipe-full stall on chatty agents. - Make agent getCwd worktree-aware using the same currentDiffType check the AI endpoints use, so agents spawn in the correct checkout. - Exclude 'agents' tab from PR context fetch guard in ReviewPanel. For provenance purposes, this commit was AI assisted. * feat(pi): mirror agent-job endpoints in Pi review server Add apps/pi-extension/server/agent-jobs.ts — a Node.js port of the Bun agent-jobs handler using child_process.spawn, res.write() SSE, and execFileSync for capability detection. Wire into serverReview.ts: route delegation, killAll in stop(), process.on("exit") guard, worktree-aware getCwd, and serverUrl late-binding. Add agent-jobs to vendor.sh for shared type vendoring. For provenance purposes, this commit was AI assisted. * fix(agents): harden agent jobs — listener leak, provider allowlist, Windows support, docs - Use process.once instead of process.on to prevent exit listener accumulation - Validate provider against known capabilities before spawning (rejects unknown providers) - Use platform-aware which/where for CLI detection on Windows - Broadcast jobs:cleared event from kill-all endpoint (was dead client code) - Document all /api/agents/* endpoints in CLAUDE.md Review Server table For provenance purposes, this commit was AI assisted. * fix(agents): SSE ref reset, jobs:cleared divergence, exit listener cleanup - Reset receivedSnapshotRef/fallbackRef on re-mount so polling fallback activates correctly on subsequent tab switches - Remove jobs:cleared broadcast from kill-all — killAll() already fires individual job:completed events, so the UI shows status transitions instead of going blank - Store exit handler ref so stop() can removeListener, preventing accumulation in long-lived hosts (OpenCode, Pi) For provenance purposes, this commit was AI assisted.	2026-03-31 10:06:18 -07:00
Michael Ramos	4627f75426	feat: external annotations API with real-time SSE (#400) ... Adds a general-purpose External Annotations API that allows external programs (linters, AI tools, security scanners) to push annotations into a live Plannotator session via HTTP, with real-time delivery over SSE. ## What's included - **Shared core** (`packages/shared/external-annotation.ts`): types, in-memory store, input validation, SSE serialization - **Server handlers**: Bun + Pi implementations with full CRUD (GET/POST/PATCH/DELETE) + SSE streaming - **Client hook** (`useExternalAnnotations`): EventSource with polling fallback, optimistic updates - **Editor integration**: two-array state model (local + external), content-aware dedup, ID-based routing - **Persistence**: source field preserved through share URLs and crash-recovery drafts - **Docs**: new Integrations category with API overview page, updated API reference ## API surface All three servers (plan, review, annotate) expose: - `GET /api/external-annotations/stream` - SSE stream - `GET /api/external-annotations` - JSON snapshot (polling fallback) - `POST /api/external-annotations` - Add annotations (single or batch) - `PATCH /api/external-annotations?id=` - Update fields - `DELETE /api/external-annotations` - Remove by id, source, or clear all For provenance purposes, this commit was AI assisted.	2026-03-29 17:18:31 -07:00
Michael Ramos	a278fdaf77	feat: plan archive browser (#369) ... * feat: plan archive browser with in-session sidebar tab (#362) Add `plannotator archive` subcommand and archive sidebar tab for browsing saved plan decisions from ~/.plannotator/plans/. Plans show approved/denied badges and timestamps. In-session mode uses the linked doc overlay pattern so users can reference old plans while reviewing a current one. - New archive server (packages/server/archive.ts) following annotate pattern - New ArchiveBrowser sidebar component, reusable in both contexts - Archive listing/parsing functions in storage.ts (reads decision snapshots) - Archive endpoints on plan server for in-session use (/api/archive/plans) - Remove dead "Other Plans" UI, projectPlans state, /api/plan/history endpoint - Fix resize handle touch area covering scrollbars in sidebar/main content - Fix sidebar tab bar overflow when narrow For provenance purposes, this commit was AI assisted. * fix: code quality sweep for plan archive - Remove `as any` cast: add "archive" to SessionInfo.mode union - Replace inline import() type with proper import for ArchivedPlan - Replace any[] with ArchivedPlan[] in fetch response types - Fix infinite re-fetch when archive is empty (use hasFetched ref) - Cache archive plan list in plan server (avoid re-scanning filesystem) - Document ResizeHandle side prop behavior - Remove redundant comment on Viewer archiveInfo prop For provenance purposes, this commit was AI assisted. * chore: remove dead marketing components Step.astro and Landing.astro are unused — landing page inlines step markup and pages use Base.astro directly. For provenance purposes, this commit was AI assisted. * fix: address code review findings for plan archive - Path traversal: use resolve() + trailing separator guard (matches reference-handlers.ts) - Thread customPath into in-session archive endpoints via query param - Sort same-day archive entries by mtime instead of title - Clear selectedArchiveFile on linked doc back to prevent badge leak - Hide archive tab in annotate mode (server doesn't serve those endpoints) - Add targetTab param to useLinkedDoc.open() to preserve calling sidebar tab - Replace mutable render variable with index-based date grouping For provenance purposes, this commit was AI assisted. * refactor: collapse standalone archive server into plan server Delete packages/server/archive.ts (187 lines) — nearly all duplicated from the plan server. Add mode:"archive" option to startPlannotatorServer instead. Fixes two bugs from code review: - handleArchiveCopy now splits on "# Plan Feedback" marker instead of bare "---", preventing truncation at horizontal rules in plan content - customPath support works in standalone archive mode (was only working in-session because the standalone server never received it) For provenance purposes, this commit was AI assisted. * refactor: extract useArchive hook from App.tsx Move archive state (archiveMode, plans, selectedFile, isLoading) and handlers (select, fetchPlans, done, copy) into a dedicated useArchive hook. Reduces App.tsx by ~75 lines and makes the archive feature self-contained. For provenance purposes, this commit was AI assisted. * feat: Pi archive parity + eliminate server duplication Move runtime-agnostic storage, draft, and project functions from packages/server/ to packages/shared/ — eliminating ~250 lines of duplicated code in Pi's server.ts. Server package becomes thin re-exports, preserving all existing import paths. Add archive mode to Pi's plan review server (mode, routes, waitForDone) and register /plannotator-archive command in the Pi extension. Consolidate ArchivedPlan type to single definition in shared/storage.ts. Simplify archive copy to include full content with feedback. For provenance purposes, this commit was AI assisted. * fix: drop -core suffix from Pi shared copies The -core suffix broke cross-file imports — storage.ts imports from ./project which didn't resolve to project-core.ts. Using the original filenames (no collision) lets relative imports work naturally. For provenance purposes, this commit was AI assisted. * fix: archive custom path bugs, disable sharing, update docs - Normalize planDir via resolve() in getPlanDir() to handle relative paths and trailing slashes in the path traversal guard - Re-fetch archive plans client-side with cookie-backed customPath so standalone archive respects the user's configured save location - Disable sharing in archive mode (read-only viewer, no need) - Remove dead /api/plan/history endpoint and listProjectPlans import from Pi extension - Remove dead /api/plan/history mock from dev-mock-api - Update CLAUDE.md and AGENTS.md: add archive flow, archive API endpoints, shared package structure, correct storage location, sidebar tab count, remove stale /api/plan/history references - Update hook server docstring from four to five modes For provenance purposes, this commit was AI assisted. * fix: empty archive shows demo content, stale viewer after customPath fetch - Clear demo markdown when archive opens with no plans (plan: "" was falsy, so setMarkdown was never called) - Remove redundant fetchPlans() from archive init — server already sends archivePlans in initial response - After fetchPlans() resolves with customPath results, auto-select and load the first plan into the viewer - Remove dead listProjectPlans re-export from server barrel For provenance purposes, this commit was AI assisted. * refactor: gitignore Pi shared copies, add @generated headers Pi's copied .ts files (storage, draft, project, feedback-templates, review-core) are build artifacts generated from packages/shared/. They looked like editable source files, leading to confusion about which file to edit. Now gitignored like the HTML copies, with @generated headers prepended by the build script. For provenance purposes, this commit was AI assisted. * fix: generate Pi shared copies in CI before tests The Pi .ts copies are now gitignored build artifacts. CI needs to generate them before running tests since server.test.ts transitively imports them via server.ts. For provenance purposes, this commit was AI assisted. * fix: generate Pi shared copies in release pipeline test job Same fix as test.yml — the Pi .ts copies are gitignored, so the test job in the release pipeline also needs to generate them before bun test. For provenance purposes, this commit was AI assisted. * fix: use block scalar in CI workflow to avoid YAML parse error The inline `run:` had a colon in the printf string that YAML interpreted as a mapping key. Switch to `run: |` block scalar. For provenance purposes, this commit was AI assisted.	2026-03-23 11:31:38 -07:00
Michael Ramos	8b8487db3e	docs: add AGENTS.md (copy of CLAUDE.md) ... Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-12 01:29:32 -07:00